Swap the resident model without restarting mavend (#250)

Loading a different gguf was a one-line edit to phraser.model_path plus a
restart. It is now an owner-triggered IPC call, off unless configured.

internal/phraser/swap.go holds the safety properties as code:

  - Never two models resident. The old llama-server is killed and reaped
    before the new one is launched. One 1.7B fits the Vega iGPU; a
    blue/green overlap would OOM the box, so it is not offered.
  - Atomic from a turn's point of view. Swap drains the in-flight turns
    (they finish on the old model), then refuses arrivals with ErrSwapping
    until the new server has answered /v1/models. No turn ever sees half a
    swap; refused turns fall back to the classifier cascade.
  - A failed load rolls back. If the new model does not start or does not
    probe, the previous one is reloaded and the call returns RolledBack
    with the error. If the rollback also fails the daemon says so and
    degrades to the classifier rather than pretending to serve.

Holders of the completion client are re-pointed, not rebuilt: llm.Client
guards its base URL and LLMPhraser.OnSwap re-points it, so the router, the
replier, the mail extractor and the memory evaluator follow the new port
without knowing a swap happened.

Reach is deliberately narrow. phraser.swap_models is an exact-match
allowlist of absolute paths a human wrote, rejected at startup otherwise,
so "swap the model" can never mean "load any file on my disk"; the running
model is always swappable back to. MethodSwapModel is AuthStepUp, the same
rung as mutating the tool allowlist, and /models gates POST through the
same stepUpOK the tools page uses. Nothing calls Swap on a timer and no
act, intent or utterance reaches it.

Vikunja #250

cmd/mavend/actions_query.go

@@ -8,6 +8,7 @@ import (
 	"strings"
 	"time"
 
+	"github.com/kami/maven/internal/crawl"
 	"github.com/kami/maven/internal/ipc"
 	"github.com/kami/maven/internal/memory"
 	"github.com/kami/maven/internal/morning"
@@ -78,6 +79,13 @@ var querySources = []querySource{
 	{"embed", (*reactiveHandler).queryEmbed},
 	{"memory", (*reactiveHandler).queryMemory},
 	{"notes", (*reactiveHandler).queryNotes},
+	// LAST before the model answers from memory, and that position is the whole
+	// design (Vikunja #259): local sources first. The model, his own notes and
+	// facts, and — once internal/kiwix is wired into this chain — the offline
+	// ZIMs all get their turn before anything touches the network. This source
+	// only claims a turn where he named a URL out loud, so it never competes
+	// with a local answer.
+	{"web", (*reactiveHandler).queryWeb},
 	{"general-knowledge", (*reactiveHandler).queryGeneral},
 }
 
@@ -371,6 +379,57 @@ func (h *reactiveHandler) queryNotes(ctx context.Context, t *queryTurn) (string,
 	return reply, true
 }
 
+// webPageContextRunes — how much of a fetched page is handed to the phraser.
+// Less than the crawler keeps: the rest of the 4096-token window belongs to the
+// prompt, the persona block and the reply.
+const webPageContextRunes = 1500
+
+// queryWeb — "посмотри https://example.org/x — что там?" (Vikunja #259).
+//
+// It claims a turn ONLY when he named a URL, which is what keeps a fallback from
+// becoming a habit: no URL, no fetch, and the model answers from what is local.
+// What leaves the box is the URL and nothing else — no note, no fact, no history
+// travels with it.
+func (h *reactiveHandler) queryWeb(ctx context.Context, t *queryTurn) (string, bool) {
+	link, ok := router.FirstURL(t.dec.Utterance)
+	if !ok {
+		return "", false
+	}
+	if h.crawler == nil {
+		// Claim rather than fall through: he asked about a specific page, and
+		// letting the model answer from the URL's spelling alone is how a small
+		// model invents a page's contents.
+		return "я не читаю страницы — это не настроено.", true
+	}
+	ctxFetch, cancel := context.WithTimeout(ctx, 30*time.Second)
+	defer cancel()
+	page, err := h.crawler.Page(ctxFetch, link)
+	if err != nil {
+		if errors.Is(err, crawl.ErrRobots) {
+			return "эта страница закрыта для чтения — robots.txt не разрешает.", true
+		}
+		log.Printf("voice: web: %v", err)
+		return "не получилось прочитать страницу.", true
+	}
+	if page.Text == "" {
+		return "страница открылась, но читать там нечего.", true
+	}
+	// The page is handed to the phraser the same way a note is: as context for
+	// the question he actually asked. She answers the question, she does not
+	// recite the page.
+	snippet := page.Title + "\n" + crawl.TrimRunes(page.Text, webPageContextRunes)
+	reply, perr := h.phraser.PhraseQuery(ctx, t.dec.Utterance, []string{snippet})
+	if perr != nil {
+		log.Printf("voice: web: phrase: %v", perr)
+	}
+	if reply == "" {
+		// No phraser (or it failed): read back the top of the page rather than
+		// pretend the fetch did not happen.
+		return "вот что на странице: " + crawl.TrimRunes(page.Text, 300), true
+	}
+	return reply, true
+}
+
 // queryGeneral — general knowledge from the phraser, the last source before
 // giving up. It always claims: either the model answers or Maven says she
 // doesn't know.

cmd/mavend/crawls.go

@@ -0,0 +1,184 @@
+// mavend/crawls.go — the driver for reading web pages (Vikunja #259,
+// docs/plans/14-web-crawler.md). The crawler is pure and lives in
+// internal/crawl; this is the impure half: the guarded fetcher, a ticker for the
+// scheduled watches, and the fact-backed dedup hashes.
+//
+// Two paths, one config block, both off unless configured:
+//
+//   - ON DEMAND — he names a URL out loud and she reads it. That is the
+//     `queryWeb` source in actions_query.go, LAST in the chain: after his
+//     memory, after the notes, and (once Kiwix is wired into the chain) after
+//     the local ZIMs. A local read costs nothing and leaks nothing; a fetch puts
+//     a URL in someone's log, so it goes last.
+//   - SCHEDULED — a watched page is re-read on its interval, and a page whose
+//     text changed is written as a note. It does NOT announce itself. Same rule
+//     as the feed poller: notes, never nudges.
+//
+// Only the URL goes out. Nothing here reads a note, a fact, the persona block or
+// the history, and internal/crawl has no access to the store at all.
+package main
+
+import (
+	"context"
+	"log"
+	"net/url"
+	"time"
+
+	"github.com/kami/maven/internal/config"
+	"github.com/kami/maven/internal/crawl"
+	"github.com/kami/maven/internal/ipc"
+	"github.com/kami/maven/internal/router"
+	"github.com/kami/maven/internal/webfetch"
+)
+
+// newCrawler builds the crawler from the `crawl` block, or returns nil when
+// there is none. Every caller checks for nil, and nil means no page is ever
+// fetched.
+func newCrawler(cfg *config.Config) *crawl.Crawler {
+	if cfg.Crawl == nil {
+		return nil
+	}
+	cc := cfg.Crawl
+
+	hosts := append([]string(nil), cc.AllowHosts...)
+	// A watched page's own host is always reachable; otherwise an allowlist and
+	// a watch list would have to be kept in sync by hand.
+	for _, w := range cc.Watches {
+		if u, err := url.Parse(w.URL); err == nil && u.Hostname() != "" {
+			hosts = append(hosts, u.Hostname())
+		}
+	}
+	// An allowlist plus on-demand is a contradiction worth logging rather than
+	// silently resolving: he asked for arbitrary pages AND for a fixed list.
+	// The allowlist wins, because it is the narrower instruction.
+	if len(hosts) > 0 && cc.OnDemand && len(cc.AllowHosts) > 0 {
+		log.Printf("crawl: allow_hosts is set, so on-demand reading is limited to those hosts")
+	}
+	ua := cc.UserAgent
+	if ua == "" {
+		ua = webfetch.DefaultUserAgent
+	}
+	fetcher := webfetch.New(webfetch.Config{
+		AllowHosts: hosts,
+		DenyHosts:  cc.DenyHosts,
+		Timeout:    time.Duration(cc.Timeout),
+		MaxBytes:   cc.MaxBytes,
+		UserAgent:  ua,
+	})
+	// The user-agent handed to the crawler is the one the fetcher sends: obeying
+	// robots rules written for a different name would be a lie.
+	return crawl.New(&crawlFetcher{f: fetcher}, crawl.Config{
+		UserAgent: ua,
+		MaxRunes:  cc.MaxRunes,
+	})
+}
+
+// onDemandCrawler returns a crawler for the answer path, or nil when on-demand
+// reading is off. The scheduled watches can be on while this is off: reading a
+// fixed list of pages on a timer and reading whatever URL is in an utterance are
+// different permissions, and the config keeps them separate.
+func onDemandCrawler(cfg *config.Config) *crawl.Crawler {
+	if cfg.Crawl == nil || !cfg.Crawl.OnDemand {
+		return nil
+	}
+	return newCrawler(cfg)
+}
+
+// crawlWorker — ticker + watcher for the scheduled half.
+type crawlWorker struct {
+	watcher  *crawl.Watcher
+	interval time.Duration
+}
+
+// crawlTickInterval — how often the worker asks what is due. Per-watch cadence
+// is the watcher's business.
+const crawlTickInterval = 15 * time.Minute
+
+// newCrawlWorker wires the scheduled crawls, or nil when nothing is watched.
+func newCrawlWorker(c *crawl.Crawler, api ipc.CoreAPI, emb router.Embedder, cfg *config.Config) *crawlWorker {
+	if c == nil || cfg.Crawl == nil || len(cfg.Crawl.Watches) == 0 {
+		return nil
+	}
+	watches := make([]crawl.WatchConfig, 0, len(cfg.Crawl.Watches))
+	for _, w := range cfg.Crawl.Watches {
+		watches = append(watches, crawl.WatchConfig{
+			Name:     w.Name,
+			URL:      w.URL,
+			Interval: time.Duration(w.Interval),
+		})
+	}
+	watcher := crawl.NewWatcher(c, watches, api, &factHashes{api: api},
+		crawlEmbedder(emb), time.Duration(cfg.Crawl.Interval))
+	if watcher == nil {
+		log.Printf("crawl: configured but nothing watchable — scheduled crawls disabled")
+		return nil
+	}
+	log.Printf("crawl: watching %d page(s), checking what is due every %s", len(watches), crawlTickInterval)
+	return &crawlWorker{watcher: watcher, interval: crawlTickInterval}
+}
+
+// run checks what is due until ctx is canceled. The first round runs
+// immediately; it writes notes only, so an early round startles nobody.
+func (w *crawlWorker) run(ctx context.Context) {
+	w.watcher.CheckDue(ctx, time.Now())
+	t := time.NewTicker(w.interval)
+	defer t.Stop()
+	for {
+		select {
+		case <-ctx.Done():
+			return
+		case now := <-t.C:
+			w.watcher.CheckDue(ctx, now)
+		}
+	}
+}
+
+// crawlFetcher adapts webfetch to crawl.Fetcher, which is the seam that keeps
+// net/http out of the crawler package.
+type crawlFetcher struct{ f *webfetch.Fetcher }
+
+func (a *crawlFetcher) Get(ctx context.Context, u string) (*crawl.Response, error) {
+	resp, err := a.f.Get(ctx, u)
+	if err != nil {
+		return nil, err
+	}
+	return &crawl.Response{URL: resp.URL, ContentType: resp.ContentType, Body: resp.Body}, nil
+}
+
+// factHashes stores each watch's last content hash as a config fact, so a
+// restart does not re-note an unchanged page. Same mechanism the feed reader
+// uses for its marks, and inspectable on /dash.
+type factHashes struct{ api ipc.CoreAPI }
+
+func hashKey(name string) string { return "crawl:hash:" + name }
+
+func (h *factHashes) LastHash(ctx context.Context, name string) (string, error) {
+	f, err := h.api.LatestFact(ctx, hashKey(name))
+	if err != nil {
+		// No hash yet is not an error: the watcher treats "" as "never read".
+		return "", nil
+	}
+	return f.Value, nil
+}
+
+func (h *factHashes) SetHash(ctx context.Context, name, hash string) error {
+	_, err := h.api.WriteFact(ctx, ipc.WriteFactReq{
+		Ts:         time.Now(),
+		Kind:       "config",
+		Key:        hashKey(name),
+		Value:      hash,
+		Source:     "poll:crawl",
+		Confidence: 1.0,
+	})
+	return err
+}
+
+// crawlEmbedder adapts router.Embedder for the watcher, embedding with
+// EmbedPassage (a page is text being searched FOR, and the e5 embedder is
+// asymmetric).
+func crawlEmbedder(emb router.Embedder) crawl.Embedder {
+	if emb == nil {
+		return nil
+	}
+	return passageEmbedder{emb}
+}

cmd/mavend/crawls_test.go

@@ -0,0 +1,186 @@
+package main
+
+import (
+	"context"
+	"net/http"
+	"net/http/httptest"
+	"strings"
+	"testing"
+
+	"github.com/kami/maven/internal/config"
+	"github.com/kami/maven/internal/crawl"
+	"github.com/kami/maven/internal/ipc"
+	"github.com/kami/maven/internal/phraser"
+	"github.com/kami/maven/internal/router"
+	"github.com/kami/maven/internal/voice"
+)
+
+// The default config reads nothing. This is the whole "off unless configured"
+// contract for the crawler, asserted at the wiring level rather than trusted.
+func TestCrawlOffByDefault(t *testing.T) {
+	cfg := &config.Config{}
+	if c := newCrawler(cfg); c != nil {
+		t.Error("newCrawler with no crawl block returned a crawler")
+	}
+	if c := onDemandCrawler(cfg); c != nil {
+		t.Error("onDemandCrawler with no crawl block returned a crawler")
+	}
+	if w := newCrawlWorker(nil, nil, nil, cfg); w != nil {
+		t.Error("newCrawlWorker with no crawl block returned a worker")
+	}
+	// Watches configured but on_demand off ⇒ the answer path still reads
+	// nothing: a timer over a fixed list is not permission for arbitrary URLs.
+	withWatch := &config.Config{Crawl: &config.CrawlConfig{
+		Watches: []config.CrawlWatchConfig{{Name: "p", URL: "https://example.org/p"}},
+	}}
+	if c := onDemandCrawler(withWatch); c != nil {
+		t.Error("onDemandCrawler honoured a watch list as on-demand permission")
+	}
+	if c := newCrawler(withWatch); c == nil {
+		t.Error("newCrawler returned nil for a configured watch")
+	}
+}
+
+// The wired fetcher must refuse a private address, because the crawler on this
+// box sits one hop from the whole homelab. Same guard the webfetch tests cover;
+// this asserts the daemon actually wires it.
+func TestCrawlerRefusesPrivateAddress(t *testing.T) {
+	srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		w.Header().Set("Content-Type", "text/html")
+		w.Write([]byte("<html><body>secret</body></html>"))
+	}))
+	defer srv.Close()
+
+	c := newCrawler(&config.Config{Crawl: &config.CrawlConfig{OnDemand: true}})
+	if c == nil {
+		t.Fatal("newCrawler returned nil for an on-demand config")
+	}
+	if _, err := c.Page(context.Background(), srv.URL); err == nil {
+		t.Fatalf("reading %s succeeded; a loopback address must be refused", srv.URL)
+	}
+}
+
+func TestFactHashesRoundTrip(t *testing.T) {
+	ctx := context.Background()
+	st := newTestStore(t)
+	h := &factHashes{api: ipc.NewStoreAPI(st)}
+
+	got, err := h.LastHash(ctx, "page")
+	if err != nil {
+		t.Fatalf("LastHash on a fresh store: %v", err)
+	}
+	if got != "" {
+		t.Errorf("LastHash = %q, want empty for a never-read page", got)
+	}
+	if err := h.SetHash(ctx, "page", "deadbeef"); err != nil {
+		t.Fatalf("SetHash: %v", err)
+	}
+	got, err = h.LastHash(ctx, "page")
+	if err != nil {
+		t.Fatalf("LastHash: %v", err)
+	}
+	if got != "deadbeef" {
+		t.Errorf("LastHash = %q, want deadbeef", got)
+	}
+	if key := hashKey("page"); key != "crawl:hash:page" {
+		t.Errorf("hashKey = %q", key)
+	}
+}
+
+// stubCrawlFetcher serves one fixed page to every URL, so queryWeb can be
+// exercised without a network or an allowlist.
+type stubCrawlFetcher struct{ body, ctype string }
+
+func (s *stubCrawlFetcher) Get(_ context.Context, u string) (*crawl.Response, error) {
+	ct := s.ctype
+	if ct == "" {
+		ct = "text/html"
+	}
+	if strings.HasSuffix(u, "/robots.txt") {
+		return &crawl.Response{URL: u, ContentType: "text/plain", Body: []byte("")}, nil
+	}
+	return &crawl.Response{URL: u, ContentType: ct, Body: []byte(s.body)}, nil
+}
+
+func buildWebHandler(c *crawl.Crawler) *reactiveHandler {
+	return &reactiveHandler{
+		replier: voice.NewStubReplier(),
+		phraser: phraser.NewStub(),
+		crawler: c,
+	}
+}
+
+func askWeb(h *reactiveHandler, q string) (string, bool) {
+	return h.queryWeb(context.Background(), &queryTurn{
+		dec: router.Decision{Intent: router.IntentQuery, Utterance: q},
+	})
+}
+
+func TestQueryWebPassesWithoutAURL(t *testing.T) {
+	h := buildWebHandler(crawl.New(&stubCrawlFetcher{body: "<html><body>x</body></html>"}, crawl.Config{}))
+	if reply, ok := askWeb(h, "почему небо синее?"); ok {
+		t.Errorf("the web source claimed a question with no URL: %q", reply)
+	}
+}
+
+// Not configured is said out loud rather than falling through, so a small model
+// never invents a page's contents from its URL.
+func TestQueryWebSaysWhenNotConfigured(t *testing.T) {
+	h := buildWebHandler(nil)
+	reply, ok := askWeb(h, "посмотри https://example.org/page")
+	if !ok {
+		t.Fatal("the web source did not claim a question with a URL")
+	}
+	if !strings.Contains(reply, "не настроено") {
+		t.Errorf("reply = %q, want the not-configured answer", reply)
+	}
+}
+
+func TestQueryWebReadsThePage(t *testing.T) {
+	h := buildWebHandler(crawl.New(&stubCrawlFetcher{
+		body: "<html><head><title>Заголовок</title></head><body><p>текст страницы</p></body></html>",
+	}, crawl.Config{}))
+	reply, ok := askWeb(h, "посмотри https://example.org/page — что там?")
+	if !ok {
+		t.Fatal("the web source did not claim a question with a URL")
+	}
+	if !strings.Contains(reply, "текст страницы") {
+		t.Errorf("reply = %q, want the page text read back", reply)
+	}
+}
+
+func TestQueryWebRefusesNonHTML(t *testing.T) {
+	h := buildWebHandler(crawl.New(&stubCrawlFetcher{
+		body: "\x00\x01binary", ctype: "application/octet-stream",
+	}, crawl.Config{}))
+	reply, ok := askWeb(h, "почитай https://example.org/blob.bin")
+	if !ok {
+		t.Fatal("the web source did not claim a question with a URL")
+	}
+	if !strings.Contains(reply, "не получилось") {
+		t.Errorf("reply = %q, want the read-failed answer", reply)
+	}
+}
+
+// robots.txt is honoured on the answer path too, and she says so instead of
+// reporting a generic failure.
+func TestQueryWebObeysRobots(t *testing.T) {
+	h := buildWebHandler(crawl.New(&robotsDenyFetcher{}, crawl.Config{}))
+	reply, ok := askWeb(h, "посмотри https://example.org/private")
+	if !ok {
+		t.Fatal("the web source did not claim a question with a URL")
+	}
+	if !strings.Contains(reply, "robots.txt") {
+		t.Errorf("reply = %q, want the robots answer", reply)
+	}
+}
+
+type robotsDenyFetcher struct{}
+
+func (robotsDenyFetcher) Get(_ context.Context, u string) (*crawl.Response, error) {
+	if strings.HasSuffix(u, "/robots.txt") {
+		return &crawl.Response{URL: u, ContentType: "text/plain",
+			Body: []byte("User-agent: *\nDisallow: /private\n")}, nil
+	}
+	return &crawl.Response{URL: u, ContentType: "text/html", Body: []byte("<html>nope</html>")}, nil
+}

cmd/mavend/mail.go

@@ -27,7 +27,6 @@ import (
 	"github.com/kami/maven/internal/config"
 	"github.com/kami/maven/internal/email"
 	"github.com/kami/maven/internal/ipc"
-	"github.com/kami/maven/internal/llm"
 	"github.com/kami/maven/internal/phraser"
 	"github.com/kami/maven/internal/store"
 )
@@ -66,7 +65,7 @@ func newMailIntake(st *store.Store, phr phraser.Phraser, cfg *config.Config) *ma
 	if timeout <= 0 {
 		timeout = config.DefaultEmailTimeout
 	}
-	ex := email.NewExtractor(llm.New(lp.BaseURL(), timeout), cfg.Email.MaxTasks, contextBlockFn(cfg, time.Now))
+	ex := email.NewExtractor(llmClientFor(lp, timeout), cfg.Email.MaxTasks, contextBlockFn(cfg, time.Now))
 	log.Printf("mail intake: enabled (max %d candidates per message, timeout %s)", cfg.Email.MaxTasks, timeout)
 	return &mailIntake{st: st, ex: ex, timeout: timeout, now: time.Now}
 }

cmd/mavend/main.go

@@ -171,6 +171,7 @@ func run(args []string) error {
 		factWorker *factEnrichmentWorker
 		evalWorker *memoryEvalWorker // nil ⇒ memory evaluation off (the default)
 		feedWkr    *feedWorker       // nil ⇒ no feed is read (the default)
+		crawlWkr   *crawlWorker      // nil ⇒ no page is watched (the default)
 	)
 
 	if !locked {
@@ -267,6 +268,7 @@ func run(args []string) error {
 		factWorker = newFactEnrichmentWorker(st, eco, time.Duration(cfg.FactEnrichmentInterval))
 		evalWorker = newMemoryEvalWorker(st, phr, cfg)
 		feedWkr = newFeedWorker(ipc.NewStoreAPI(st), embedderOf(voiceW), cfg)
+		crawlWkr = newCrawlWorker(newCrawler(cfg), ipc.NewStoreAPI(st), embedderOf(voiceW), cfg)
 
 		coreAPI = &daemonAPI{
 			CoreAPI:          ipc.NewStoreAPI(st),
@@ -327,6 +329,7 @@ func run(args []string) error {
 	// ipc.MethodIngestMail reports ErrUnknownMethod.
 	if !locked {
 		wireMailIntake(srv, st, phr, cfg)
+		wireModelSwap(srv, phr, cfg)
 	}
 
 	// WrapKeyFn — wraps the env key with a passkey credential public key and
@@ -454,6 +457,7 @@ func run(args []string) error {
 			factWorker = newFactEnrichmentWorker(st, eco, time.Duration(cfg.FactEnrichmentInterval))
 			evalWorker = newMemoryEvalWorker(st, phr, cfg)
 			feedWkr = newFeedWorker(ipc.NewStoreAPI(st), embedderOf(voiceW), cfg)
+			crawlWkr = newCrawlWorker(newCrawler(cfg), ipc.NewStoreAPI(st), embedderOf(voiceW), cfg)
 
 			// Swap the CoreAPI from the locked placeholder to the real store adapter.
 			newAPI := &daemonAPI{
@@ -468,6 +472,7 @@ func run(args []string) error {
 			srv.SetAPI(newAPI)
 			srv.Check = (&auth.Gate{Enrollment: auth.NewFloorEnrollment(), Session: passkeySess}).Check
 			wireMailIntake(srv, st, phr, cfg)
+			wireModelSwap(srv, phr, cfg)
 
 			// Start voice server.
 			if voiceW != nil {
@@ -506,6 +511,13 @@ func run(args []string) error {
 				}()
 			}
 
+			// Start the watched-page crawls (nil unless configured).
+			if crawlWkr != nil {
+				go func() {
+					crawlWkr.run(ctx)
+				}()
+			}
+
 			dl.unlock()
 			log.Printf("mavend: unlocked via passkey assertion")
 			return nil
@@ -558,6 +570,13 @@ func run(args []string) error {
 				feedWkr.run(ctx)
 			}()
 		}
+		if crawlWkr != nil {
+			wg.Add(1)
+			go func() {
+				defer wg.Done()
+				crawlWkr.run(ctx)
+			}()
+		}
 	}
 
 	<-ctx.Done()

cmd/mavend/memoryeval.go

@@ -16,7 +16,6 @@ import (
 	"time"
 
 	"github.com/kami/maven/internal/config"
-	"github.com/kami/maven/internal/llm"
 	"github.com/kami/maven/internal/memeval"
 	"github.com/kami/maven/internal/phraser"
 	"github.com/kami/maven/internal/store"
@@ -50,7 +49,7 @@ func newMemoryEvalWorker(st *store.Store, phr phraser.Phraser, cfg *config.Confi
 	}
 	// A generous per-request timeout: this is a long prompt to a Thinking model
 	// and nobody is waiting on the answer.
-	client := llm.New(lp.BaseURL(), 5*time.Minute)
+	client := llmClientFor(lp, 5*time.Minute)
 	ev := memeval.NewEvaluator(st, st, client, memeval.Config{
 		MaxItems:      cfg.MemoryEval.MaxItems,
 		MinConfidence: cfg.MemoryEval.MinConfidence,

cmd/mavend/modelswap.go

@@ -0,0 +1,113 @@
+package main
+
+import (
+	"context"
+	"fmt"
+	"log"
+	"path/filepath"
+	"time"
+
+	"github.com/kami/maven/internal/config"
+	"github.com/kami/maven/internal/ipc"
+	"github.com/kami/maven/internal/llm"
+	"github.com/kami/maven/internal/phraser"
+)
+
+// Swapping the resident model while the daemon runs (Vikunja #250).
+//
+// Off unless configured: with no phraser.swap_models allowlist the two IPC
+// methods are never wired, so they answer ErrUnknownMethod. When it is wired the
+// swap method is AuthStepUp (internal/auth), which means an authed human surface
+// only — there is no act, no intent and no timer that reaches it. The daemon
+// never decides to change its own brain.
+//
+// The allowlist is exact-match against paths a human wrote in mavend.json. The
+// request carries a path and llama-server is started with it as `-m`, so
+// anything looser would turn "swap the model" into "load any file on my disk".
+func wireModelSwap(srv *ipc.Server, phr phraser.Phraser, cfg *config.Config) {
+	if cfg.Phraser == nil || len(cfg.Phraser.SwapModels) == 0 {
+		return
+	}
+	lp, ok := phr.(*phraser.LLMPhraser)
+	if !ok {
+		log.Printf("model swap: phraser.swap_models is set but there is no llama-server phraser — swap disabled")
+		return
+	}
+	allowed := map[string]bool{}
+	for _, m := range cfg.Phraser.SwapModels {
+		allowed[filepath.Clean(m)] = true
+	}
+	// The configured model is always swappable back to, listed or not: the way
+	// out of a bad swap must not depend on remembering to allowlist the model
+	// you are already running.
+	allowed[filepath.Clean(cfg.Phraser.ModelPath)] = true
+
+	srv.SwapModelFn = func(ctx context.Context, req ipc.SwapModelReq) (ipc.SwapModelResp, error) {
+		path := filepath.Clean(req.ModelPath)
+		if !allowed[path] {
+			log.Printf("model swap: REFUSED %q — not in phraser.swap_models", req.ModelPath)
+			return ipc.SwapModelResp{}, fmt.Errorf("%w: %q is not in phraser.swap_models", ipc.ErrForbidden, req.ModelPath)
+		}
+		res, err := lp.Swap(ctx, phraser.SwapSpec{
+			ModelPath:  path,
+			NGpuLayers: req.NGpuLayers,
+			NCtx:       req.NCtx,
+		})
+		resp := ipc.SwapModelResp{
+			Model:      res.Model,
+			ModelPath:  res.ModelPath,
+			BaseURL:    res.BaseURL,
+			RolledBack: res.RolledBack,
+			TookMs:     res.Took.Milliseconds(),
+		}
+		if err != nil {
+			// A rolled-back swap is a failure that left a working daemon behind.
+			// Both halves matter to the caller, so the response is filled in even
+			// though the error is returned.
+			log.Printf("model swap: %v", err)
+			return resp, err
+		}
+		return resp, nil
+	}
+
+	srv.ModelStatusFn = func(ctx context.Context) (ipc.ModelStatusResp, error) {
+		path, ngl, nctx := lp.LiveModel()
+		base := lp.BaseURL()
+		resp := ipc.ModelStatusResp{
+			ModelPath:  path,
+			BaseURL:    base,
+			NGpuLayers: ngl,
+			NCtx:       nctx,
+			Swappable:  cfg.Phraser.SwapModels,
+		}
+		if base == "" {
+			resp.Model = llm.UnknownModel
+			return resp, nil
+		}
+		id, err := llm.ModelID(ctx, base)
+		if err != nil {
+			// Report the honest "I could not confirm it" rather than echoing the
+			// configured filename as if the server had said it.
+			resp.Model = llm.UnknownModel
+			return resp, nil
+		}
+		resp.Model = id
+		return resp, nil
+	}
+
+	log.Printf("model swap: enabled, %d allowlisted model(s) — step-up required", len(cfg.Phraser.SwapModels))
+}
+
+// llmClientFor builds a completion client on the phraser's llama-server and
+// keeps it pointed at the right one across a model swap.
+//
+// Without the OnSwap registration every holder of a base URL — the LLM router,
+// the replier, the mail extractor, the memory evaluator — would keep talking to
+// the port of a server that no longer exists, and the daemon would degrade to
+// the classifier permanently after the first swap. The client is re-pointed, not
+// rebuilt, so nothing that holds it has to know a swap happened.
+func llmClientFor(lp *phraser.LLMPhraser, timeout time.Duration) *llm.Client {
+	c := llm.New(lp.BaseURL(), timeout)
+	lp.OnSwap(func(base string) { c.SetBaseURL(base) })
+	return c
+}

cmd/mavend/voice.go

@@ -52,6 +52,7 @@ import (
 	"time"
 
 	"github.com/kami/maven/internal/audio"
+	"github.com/kami/maven/internal/crawl"
 	"github.com/kami/maven/internal/dialogue"
 	"github.com/kami/maven/internal/ipc"
 	"github.com/kami/maven/internal/memory"
@@ -82,6 +83,10 @@ type reactiveHandler struct {
 	replier  voice.Replier
 	now      func() time.Time
 
+	// crawler reads a web page he names out loud (queryWeb). nil ⇒ on-demand
+	// page reading is off, which is the default: no `crawl` block, no fetch.
+	crawler *crawl.Crawler
+
 	// feedsOn — whether any RSS feed is configured (config.Feeds). It changes
 	// only what she SAYS when asked and nothing is there: "ленты не настроены"
 	// instead of "ничего нового", which are different truths.

cmd/mavend/voicewire.go

@@ -148,7 +148,9 @@ func wireVoice(cfg *config.Config, coreAPI ipc.CoreAPI, phr phraser.Phraser, mem
 	// The replier uses the same llama-server as the phraser.
 	var llmClient *llm.Client
 	if lp, ok := phr.(*phraser.LLMPhraser); ok {
-		llmClient = llm.New(lp.BaseURL(), 60*time.Second)
+		// llmClientFor, not llm.New: this client must follow the phraser onto
+		// the new llama-server when the resident model is swapped (Vikunja #250).
+		llmClient = llmClientFor(lp, 60*time.Second)
 	}
 	// ----- router (the cascade; floor examples seed the classifier) -----
 	// The act matcher's allowlist is exactly the enabled tool names — the
@@ -201,17 +203,20 @@ func wireVoice(cfg *config.Config, coreAPI ipc.CoreAPI, phr phraser.Phraser, mem
 
 	// ----- the handler (the reactive path; closes over stt / tts / router / coreAPI / memory) -----
 	h := &reactiveHandler{
-		stt:              transcriber,
-		tts:              synthesizer,
-		router:           rtr,
-		embedder:         emb,
-		api:              coreAPI,
-		tools:            exec,
-		matcher:          matcher,
-		replier:          replier,
-		phraser:          phr,
-		now:              time.Now,
-		feedsOn:          cfg.Feeds != nil,
+		stt:      transcriber,
+		tts:      synthesizer,
+		router:   rtr,
+		embedder: emb,
+		api:      coreAPI,
+		tools:    exec,
+		matcher:  matcher,
+		replier:  replier,
+		phraser:  phr,
+		now:      time.Now,
+		feedsOn:  cfg.Feeds != nil,
+		// nil unless `crawl.on_demand` is on: reading a page he names is a
+		// capability, and capabilities are off unless configured.
+		crawler:          onDemandCrawler(cfg),
 		weatherProvider:  weatherProvider,
 		weatherLocation:  weatherLocation,
 		memStore:         memStore,

cmd/mavweb/main.go

@@ -124,6 +124,7 @@ var sidebarSections = []struct {
 		Label: "Settings",
 		Pages: []struct{ Label, URL, Key string }{
 			{Label: "Tools", URL: "/tools", Key: "tools"},
+			{Label: "Model", URL: "/models", Key: "models"},
 			{Label: "Passkey", URL: "/auth/passkey", Key: "passkey"},
 		},
 	},
@@ -191,6 +192,8 @@ func pageIcon(key string) string {
 		return `<svg class=icon width="14" height="14"><use href="/ethos-icons.svg#i-grid"/></svg>`
 	case "tools":
 		return `<svg class=icon width="14" height="14"><use href="/ethos-icons.svg#i-settings"/></svg>`
+	case "models":
+		return `<svg class=icon width="14" height="14"><use href="/ethos-icons.svg#i-wave"/></svg>`
 	case "passkey":
 		return `<svg class=icon width="14" height="14"><use href="/ethos-icons.svg#i-lock"/></svg>`
 	default:
@@ -225,6 +228,8 @@ func pageTitle(key string) string {
 		return "Ecosystem"
 	case "tools":
 		return "Tools"
+	case "models":
+		return "Resident Model"
 	case "passkey":
 		return "Passkey"
 	default:
@@ -479,6 +484,12 @@ func main() {
 	mux.HandleFunc("/routines", func(w http.ResponseWriter, r *http.Request) {
 		handleRoutines(w, r, core, stepUpSession, *requireStepUp)
 	})
+	// /models — the resident-model surface (Vikunja #250). Same step-up gate as
+	// /tools, and for a comparable reason: which model is loaded decides how every
+	// utterance is routed and how every reply is worded. GET is read-only.
+	mux.HandleFunc("/models", func(w http.ResponseWriter, r *http.Request) {
+		handleModels(w, r, core, stepUpSession, *requireStepUp)
+	})
 
 	// State-changing routes on this server, and their gate (Vikunja #317):
 	//

cmd/mavweb/models.go

@@ -0,0 +1,145 @@
+package main
+
+import (
+	"context"
+	"errors"
+	"html/template"
+	"log"
+	"net/http"
+	"strconv"
+	"strings"
+
+	"github.com/kami/maven/internal/ipc"
+	"github.com/kami/maven/internal/webauthn"
+)
+
+// The resident-model surface (Vikunja #250).
+//
+// GET shows which model llama-server actually has loaded and which files the
+// daemon is configured to allow. POST swaps to one of them, behind the same
+// step-up gate as POST /tools: the loaded model decides how every utterance is
+// routed and how every reply is worded, so it is an owner action.
+//
+// There is nothing on this page Maven can press. The swap is an IPC method rated
+// AuthStepUp in internal/auth, unreachable from an act, an intent or a timer.
+
+// modelController — the two non-CoreAPI methods this page needs. *ipc.Client
+// satisfies it; a core without a swap allowlist answers ErrUnknownMethod, which
+// the page renders as "not configured" rather than an error.
+type modelController interface {
+	ModelStatus(ctx context.Context) (ipc.ModelStatusResp, error)
+	SwapModel(ctx context.Context, req ipc.SwapModelReq) (ipc.SwapModelResp, error)
+}
+
+var modelsTmpl = template.Must(template.New("models").Funcs(shellFuncs()).Parse(shellTopHTML + modelsHTML + shellBottomHTML))
+
+const modelsHTML = `{{template "shellTop" "models"}}
+<h1>Resident model</h1>
+<p class=hint>swapping requires step-up — <a href=/auth/passkey>assert a passkey</a> first. The old model is unloaded before the new one is loaded (one model fits the iGPU at a time), so turns during the load are refused and fall back to the classifier.</p>
+{{if .Msg}}<div class="msg msg-ok">{{.Msg}}</div>{{end}}
+{{if .Err}}<div class="msg msg-err">{{.Err}}</div>{{end}}
+{{if .Off}}
+<section class=card>
+<h2 class=card-title>swap not configured</h2>
+<p class=hint>this core has no <code>phraser.swap_models</code> allowlist, so there is nothing to swap to. Add the gguf paths you allow to <code>deploy/mavend.json</code> and restart once.</p>
+</section>
+{{else}}
+<section class=card>
+<h2 class=card-title>loaded now</h2>
+<div class=scroll><table>
+<tr><th>model</th><td><code>{{.Status.Model}}</code></td></tr>
+<tr><th>file</th><td><code>{{.Status.ModelPath}}</code></td></tr>
+<tr><th>server</th><td><code>{{.Status.BaseURL}}</code></td></tr>
+<tr><th>n_ctx</th><td>{{.Status.NCtx}}</td></tr>
+<tr><th>n_gpu_layers</th><td>{{.Status.NGpuLayers}}</td></tr>
+</table></div>
+<p class=hint>the model name is what llama-server reports for itself, not what the config says it should be.</p>
+</section>
+<section class=card>
+<h2 class=card-title>allowed models <span class=badge>{{len .Status.Swappable}}</span></h2>
+{{if .Status.Swappable}}<div class=scroll><table><tr><th>file</th><th></th></tr>
+{{range .Status.Swappable}}<tr><td><code>{{.}}</code></td>
+<td><form method=post action=/models class=inline-form>
+<input type=hidden name=model_path value="{{.}}">
+<button class=btn>load this one</button></form></td></tr>{{end}}
+</table></div>
+{{else}}<div class=empty><div>no models allowlisted</div></div>{{end}}
+</section>
+{{end}}
+{{template "shellBottom"}}`
+
+type modelsPage struct {
+	Msg    string
+	Err    string
+	Off    bool
+	Status ipc.ModelStatusResp
+}
+
+// handleModels renders the model surface (GET) and applies a swap (POST).
+//
+// A failed swap is reported as a failure with the model that is still serving
+// named, because that is the state the operator needs: the daemon rolled back
+// and is answering turns, it just is not answering them with what he asked for.
+func handleModels(w http.ResponseWriter, r *http.Request, core ipc.CoreAPI, session *webauthn.PasskeySession, requireStepUp bool) {
+	if core == nil {
+		http.Error(w, "models disabled (no -core)", http.StatusServiceUnavailable)
+		return
+	}
+	mc, ok := core.(modelController)
+	if !ok {
+		http.Error(w, "models unavailable: core connection does not support model swap", http.StatusServiceUnavailable)
+		return
+	}
+	ctx := r.Context()
+	page := modelsPage{}
+
+	if r.Method == http.MethodPost {
+		if !stepUpOK(session, requireStepUp) {
+			http.Error(w, "step-up required: assert a passkey first", http.StatusForbidden)
+			return
+		}
+		path := strings.TrimSpace(r.FormValue("model_path"))
+		if path == "" {
+			http.Error(w, "model_path required", http.StatusBadRequest)
+			return
+		}
+		req := ipc.SwapModelReq{ModelPath: path}
+		if v, err := strconv.Atoi(r.FormValue("n_ctx")); err == nil {
+			req.NCtx = v
+		}
+		res, err := mc.SwapModel(ctx, req)
+		switch {
+		case err == nil:
+			page.Msg = "loaded " + res.Model + " (" + strconv.FormatInt(res.TookMs, 10) + "ms)"
+			log.Printf("models: swapped to %s (%s) in %dms", res.ModelPath, res.Model, res.TookMs)
+		case errors.Is(err, ipc.ErrForbidden):
+			http.Error(w, "refused: that model is not in phraser.swap_models, or step-up was not asserted", http.StatusForbidden)
+			return
+		case errors.Is(err, ipc.ErrUnknownMethod):
+			http.Error(w, "swap not configured on this core", http.StatusServiceUnavailable)
+			return
+		case res.RolledBack:
+			page.Err = "swap failed, rolled back to " + res.Model + " — she is still answering, with the old model"
+			log.Printf("models: swap to %s failed, rolled back: %v", path, err)
+		default:
+			page.Err = "swap failed: " + err.Error()
+			log.Printf("models: swap to %s failed: %v", path, err)
+		}
+	}
+
+	st, err := mc.ModelStatus(ctx)
+	if err != nil {
+		if errors.Is(err, ipc.ErrUnknownMethod) {
+			page.Off = true
+		} else {
+			log.Printf("models: status: %v", err)
+			http.Error(w, "core read failed", http.StatusBadGateway)
+			return
+		}
+	}
+	page.Status = st
+	w.Header().Set("Content-Type", "text/html; charset=utf-8")
+	if err := modelsTmpl.Execute(w, page); err != nil {
+		log.Printf("models render: %v", err)
+	}
+}

cmd/mavweb/models_test.go

@@ -0,0 +1,150 @@
+package main
+
+import (
+	"context"
+	"net/http"
+	"net/http/httptest"
+	"strings"
+	"testing"
+
+	"github.com/kami/maven/internal/ipc"
+	"github.com/kami/maven/internal/webauthn"
+)
+
+// fakeModelCore is a core that supports the two model methods. It records what
+// the page asked for, so the tests can assert the gate rather than the HTML.
+type fakeModelCore struct {
+	ipc.UnimplementedCoreAPI
+
+	status    ipc.ModelStatusResp
+	statusErr error
+
+	swapResp ipc.SwapModelResp
+	swapErr  error
+	swapped  []ipc.SwapModelReq
+}
+
+func (f *fakeModelCore) ModelStatus(ctx context.Context) (ipc.ModelStatusResp, error) {
+	return f.status, f.statusErr
+}
+
+func (f *fakeModelCore) SwapModel(ctx context.Context, req ipc.SwapModelReq) (ipc.SwapModelResp, error) {
+	f.swapped = append(f.swapped, req)
+	return f.swapResp, f.swapErr
+}
+
+func modelsGET(t *testing.T, core ipc.CoreAPI) *httptest.ResponseRecorder {
+	t.Helper()
+	w := httptest.NewRecorder()
+	handleModels(w, httptest.NewRequest(http.MethodGet, "/models", nil), core, nil, false)
+	return w
+}
+
+func modelsPOST(t *testing.T, core ipc.CoreAPI, session *webauthn.PasskeySession, requireStepUp bool, path string) *httptest.ResponseRecorder {
+	t.Helper()
+	r := httptest.NewRequest(http.MethodPost, "/models", strings.NewReader("model_path="+path))
+	r.Header.Set("Content-Type", "application/x-www-form-urlencoded")
+	w := httptest.NewRecorder()
+	handleModels(w, r, core, session, requireStepUp)
+	return w
+}
+
+func TestModels_GETShowsTheLoadedModelAndTheAllowlist(t *testing.T) {
+	core := &fakeModelCore{status: ipc.ModelStatusResp{
+		Model:     "Qwen3-1.7B-UD-Q4_K_XL",
+		ModelPath: "/opt/maven/models/llm/qwen3.gguf",
+		BaseURL:   "http://127.0.0.1:18099",
+		NCtx:      4096,
+		Swappable: []string{"/opt/maven/models/llm/qwen3.gguf", "/opt/maven/models/llm/qwen3-cpt.gguf"},
+	}}
+	w := modelsGET(t, core)
+	if w.Code != http.StatusOK {
+		t.Fatalf("GET /models = %d; want 200", w.Code)
+	}
+	body := w.Body.String()
+	for _, want := range []string{"Qwen3-1.7B-UD-Q4_K_XL", "qwen3-cpt.gguf", "4096"} {
+		if !strings.Contains(body, want) {
+			t.Errorf("page does not mention %q", want)
+		}
+	}
+	if len(core.swapped) != 0 {
+		t.Errorf("a GET swapped the model: %v", core.swapped)
+	}
+}
+
+func TestModels_POSTRequiresStepUpWhenFailingClosed(t *testing.T) {
+	// No WebAuthn configured (nil session) + -require-stepup ⇒ deny, exactly
+	// like POST /tools. Nothing reaches core.
+	core := &fakeModelCore{}
+	w := modelsPOST(t, core, nil, true, "/opt/maven/models/llm/qwen3.gguf")
+	if w.Code != http.StatusForbidden {
+		t.Fatalf("POST /models without assertable step-up = %d; want 403", w.Code)
+	}
+	if len(core.swapped) != 0 {
+		t.Fatalf("a denied POST still called SwapModel: %v", core.swapped)
+	}
+}
+
+func TestModels_POSTSwapsAndReportsTheModelThatAnswered(t *testing.T) {
+	core := &fakeModelCore{
+		swapResp: ipc.SwapModelResp{Model: "qwen3-cpt", ModelPath: "/m/cpt.gguf", TookMs: 4200},
+		status:   ipc.ModelStatusResp{Model: "qwen3-cpt", ModelPath: "/m/cpt.gguf"},
+	}
+	w := modelsPOST(t, core, nil, false, "/m/cpt.gguf")
+	if w.Code != http.StatusOK {
+		t.Fatalf("POST /models = %d; want 200", w.Code)
+	}
+	if len(core.swapped) != 1 || core.swapped[0].ModelPath != "/m/cpt.gguf" {
+		t.Fatalf("SwapModel calls = %v; want one for /m/cpt.gguf", core.swapped)
+	}
+	if !strings.Contains(w.Body.String(), "loaded qwen3-cpt") {
+		t.Errorf("page does not report which model was loaded:\n%s", w.Body.String())
+	}
+}
+
+func TestModels_RolledBackSwapSaysSheIsStillAnswering(t *testing.T) {
+	core := &fakeModelCore{
+		swapResp: ipc.SwapModelResp{Model: "qwen3", ModelPath: "/m/old.gguf", RolledBack: true},
+		swapErr:  errBrokenModel{},
+		status:   ipc.ModelStatusResp{Model: "qwen3", ModelPath: "/m/old.gguf"},
+	}
+	w := modelsPOST(t, core, nil, false, "/m/cpt.gguf")
+	if w.Code != http.StatusOK {
+		t.Fatalf("POST /models after a rollback = %d; want 200 with the failure rendered", w.Code)
+	}
+	body := w.Body.String()
+	if !strings.Contains(body, "rolled back to qwen3") {
+		t.Errorf("page does not say it rolled back:\n%s", body)
+	}
+}
+
+func TestModels_RefusedPathIs403(t *testing.T) {
+	core := &fakeModelCore{swapErr: ipc.ErrForbidden}
+	w := modelsPOST(t, core, nil, false, "/etc/passwd")
+	if w.Code != http.StatusForbidden {
+		t.Fatalf("POST /models with a non-allowlisted path = %d; want 403", w.Code)
+	}
+}
+
+func TestModels_UnconfiguredCoreRendersOff(t *testing.T) {
+	core := &fakeModelCore{statusErr: ipc.ErrUnknownMethod}
+	w := modelsGET(t, core)
+	if w.Code != http.StatusOK {
+		t.Fatalf("GET /models against a core without the swap = %d; want 200", w.Code)
+	}
+	if !strings.Contains(w.Body.String(), "swap not configured") {
+		t.Errorf("page does not say the capability is off:\n%s", w.Body.String())
+	}
+}
+
+func TestModels_CoreWithoutTheMethodsIs503(t *testing.T) {
+	// An in-process CoreAPI (no swap methods) must not 500 the page.
+	w := modelsGET(t, ipc.UnimplementedCoreAPI{})
+	if w.Code != http.StatusServiceUnavailable {
+		t.Fatalf("GET /models on a core without the methods = %d; want 503", w.Code)
+	}
+}
+
+type errBrokenModel struct{}
+
+func (errBrokenModel) Error() string { return "llm: server did not start" }

deploy/README.md

@@ -67,6 +67,36 @@ What it does and does not do:
 - how far each feed was read is stored as a config fact `rss:latest:<name>`, so
   a restart does not re-note yesterday's headlines.
 
+### Reading a page (`crawl`, also off by default)
+
+There is no `crawl` block either, so no page is fetched. Two halves, separately
+switched:
+
+```json
+"crawl": {
+  "on_demand": true,
+  "interval": "6h",
+  "max_runes": 4000,
+  "watches": [
+    { "name": "changelog", "url": "https://example.org/changelog", "interval": "12h" }
+  ]
+}
+```
+
+- `on_demand` lets her read a page he names in the utterance: "посмотри
+  https://example.org/x — что там?". The page becomes context for his question,
+  and only the URL leaves the box. Without a URL nothing is fetched, so this is
+  a fallback and not a habit;
+- `watches` re-reads a fixed list on its interval and writes a note when the
+  text changed. Like the feeds, it announces nothing;
+- the answer path sits **last** in the query chain, behind his memory, his notes
+  and (once wired) the local Kiwix ZIMs. A local read costs nothing;
+- `robots.txt` is fetched first and obeyed with no override; a `Disallow` is a
+  refusal she says out loud. `Crawl-delay` is honoured;
+- same guarded fetcher as the feeds: allowlist/denylist, no private addresses,
+  size cap, redirect cap, timeout, one request per host per second;
+- dedup state is the config fact `crawl:hash:<name>`.
+
 ## Not yet verified / host-dependent
 
 This stack is correct-by-construction but has **not been build-tested here**

docs/plans/14-web-crawler.md

@@ -28,3 +28,41 @@
 7. Add IPC methods `MethodTriggerCrawl(name)`, `MethodListCrawls`, `MethodGetCrawlResult(name)`
 8. Add `crawls` block to `config.Config` and `deploy/mavend.json`
 9. Test with a static HTML page — verify extraction matches expected values, verify scheduling fires correctly
+
+## Shipped 2026-08-01 (#259)
+
+Built as `internal/crawl` (pure: robots, extraction, watcher) plus
+`cmd/mavend/crawls.go` (fetcher, ticker, dedup facts), on top of the guarded
+`internal/webfetch` door added with the feed reader (#258). Off unless
+configured, in two separately-switched halves: `crawl.on_demand` for a URL he
+names, `crawl.watches` for a scheduled re-read.
+
+**Limits are code, not documentation** (`internal/webfetch`, tested one test per
+limit): host allowlist/denylist, no private addresses (loopback, RFC1918 —
+hence the LAN and the `10.42.0.0/24` wg range —, link-local incl. cloud
+metadata, CGNAT, v6 ULA) enforced in the dialer's `Control` hook so DNS
+rebinding and every redirect hop are covered, response size cap, redirect cap,
+timeout, one request per host per second. `robots.txt` is fetched first, cached
+per host, and a `Disallow` is refused with no override.
+
+Deliberate deviations from the plan above:
+
+- **No CSS selectors and no LLM structured extraction** (steps 2). The output is
+  plaintext handed to the phraser as context for the question he asked. A 1.7B
+  extracting a JSON price table from 4000 runes is a worse bet than reading, and
+  `goquery` is not vendored.
+- **No `crawl` act verb and no new IPC methods** (steps 5, 7). Reading a page is
+  a query source (`queryWeb` in `actions_query.go`, last in the chain, behind
+  Kiwix once that is wired), not an action he commands. Nothing needs a new wire
+  method to work.
+- **Notes, not facts.** A page's text is not a fact about him. Only the dedup
+  hash is a fact (`crawl:hash:<name>`, kind `config`, source `poll:crawl`).
+- **Nothing is dispatched.** A changed page writes a note; it does not nudge.
+  Not a nag.
+- **No `/tools` crawl history page.** The notes and the hash facts are already
+  visible on `/dash`.
+
+**No new dependency.** The vendored tree has no `x/net/html`, no `goquery` and
+no `temoto/robotstxt`, so robots parsing and HTML-to-text are stdlib
+(`regexp`, `html`) — RE2 has no backreferences, hence the `pairsRE` builder in
+`extract.go`.

internal/auth/auth_test.go

@@ -393,3 +393,26 @@ func mustWriteFactParams(source string) []byte {
 	}
 	return b
 }
+
+// TestRequirement_SwapModel — loading a different resident model is an owner
+// action at the same rung as mutating the tool allowlist: it decides how every
+// utterance is routed and how every reply is worded. The read side is not.
+func TestRequirement_SwapModel(t *testing.T) {
+	if got := Requirement(ipc.MethodSwapModel); got != AuthStepUp {
+		t.Errorf("SwapModel authority = %v; want AuthStepUp", got)
+	}
+	if got := Requirement(ipc.MethodModelStatus); got != AuthRead {
+		t.Errorf("ModelStatus authority = %v; want AuthRead", got)
+	}
+	// A surface that cannot carry a passkey gesture cannot swap the model, no
+	// matter what it is enrolled as — this is the "never through voice" property.
+	voice := Scope{Surface: SurfaceVoice, Module: "voice", SourceScope: []string{"*"}}
+	if err := Can(ipc.MethodSwapModel, voice, nil); !errors.Is(err, ErrForbidden) {
+		t.Errorf("voice swapping the model = %v; want ErrForbidden", err)
+	}
+	// And with no step-up session asserted, the gate refuses even a capable surface.
+	noSession := &Gate{Enrollment: NewFloorEnrollment()}
+	if err := noSession.Check(context.Background(), ipc.MethodSwapModel, nil); !errors.Is(err, ipc.ErrForbidden) {
+		t.Errorf("SwapModel with no asserted step-up = %v; want ErrForbidden", err)
+	}
+}

internal/auth/policy.go

@@ -53,6 +53,13 @@ func Requirement(m ipc.Method) Authority {
 		// asserted — never a module or the voice/chat path. maven can propose
 		// (MethodProposeTool, no step-up: she has no passkey) but never en/disable.
 		return AuthStepUp
+	case ipc.MethodSwapModel:
+		// Swapping the resident model changes what routes every utterance and
+		// what words every reply. It is the owner's call, from a surface that can
+		// carry a passkey gesture — the same rung as mutating the tool allowlist,
+		// and for the same reason: nothing Maven says or does may reach it.
+		// MethodModelStatus is only the read side, so it stays at AuthRead.
+		return AuthStepUp
 	case ipc.MethodWriteFact:
 		return AuthWrite
 	case ipc.MethodAssertStepUp:
@@ -79,7 +86,10 @@ func Requirement(m ipc.Method) Authority {
 		// produce: candidate tasks and nothing else. It cannot write a fact, set a
 		// reminder, or touch the tool allowlist, so a compromised mail reader can
 		// at worst put junk on a review page he clears in one click.
-		ipc.MethodIngestMail:
+		ipc.MethodIngestMail,
+		// The read side of the model swap: which model is resident, which ones are
+		// allowlisted. It loads nothing and changes nothing.
+		ipc.MethodModelStatus:
 		return AuthRead
 	}
 	// Unknown method ⇒ AuthRead, but ipc.dispatch returns ErrUnknownMethod

internal/config/config.go

@@ -161,6 +161,10 @@ type Config struct {
 	// the weather and telegram. See FeedsConfig.
 	Feeds *FeedsConfig `json:"feeds,omitempty"`
 
+	// Crawl — reading a web page (Vikunja #259). nil / absent ⇒ Maven never
+	// fetches a page: not on request, not on a schedule. See CrawlConfig.
+	Crawl *CrawlConfig `json:"crawl,omitempty"`
+
 	// Praxis — the ecosystem attention-state service. When configured, maven
 	// calls the Praxis HTTP tools API for attention listing and item lifecycle.
 	// Maven never touches Praxis's database directly (ecosystem invariant: no
@@ -454,6 +458,62 @@ type FeedSourceConfig struct {
 	Exclude  []string `json:"exclude,omitempty"`  // drop items containing any of these
 }
 
+// CrawlConfig — the web crawler (Vikunja #259, docs/plans/14-web-crawler.md).
+//
+// Absent ⇒ off, and off means no page is ever fetched. Present with neither
+// `on_demand` nor a `watches` entry is also off: there would be nothing to do.
+//
+// The crawler is the LAST place an answer is looked for, behind the model, his
+// own memory and the local Kiwix ZIMs. That ordering lives in the query-source
+// chain (cmd/mavend/actions_query.go), not here, but it is the reason this block
+// is small: it is a fallback, not a search engine.
+//
+// Only the URL leaves the box. His notes, facts, persona block and history are
+// never part of a request — the crawler package cannot even read the store.
+type CrawlConfig struct {
+	// OnDemand — may he ask her to read a page he names out loud
+	// ("посмотри https://… — что там пишут?"). false ⇒ the on-demand answer
+	// source stays off and only the watches below run.
+	OnDemand bool `json:"on_demand,omitempty"`
+
+	// Watches — pages re-read on a schedule. A page whose text changed is
+	// written as a note (source "crawl:<name>"); nothing is announced.
+	Watches []CrawlWatchConfig `json:"watches,omitempty"`
+
+	// Interval — default watch cadence. 0 ⇒ crawl.DefaultWatchInterval (6h).
+	Interval Duration `json:"interval,omitempty"`
+
+	// AllowHosts — when set, the ONLY hosts the crawler may reach (subdomains
+	// included). Watched pages' own hosts are added automatically. Setting this
+	// is how "she may read the arch wiki and nothing else" is expressed.
+	AllowHosts []string `json:"allow_hosts,omitempty"`
+
+	// DenyHosts — never reachable, checked first. Private addresses do not need
+	// to be listed: they are refused unconditionally (see internal/webfetch).
+	DenyHosts []string `json:"deny_hosts,omitempty"`
+
+	// UserAgent — sent on every request AND matched against robots.txt groups.
+	// Empty ⇒ webfetch.DefaultUserAgent.
+	UserAgent string `json:"user_agent,omitempty"`
+
+	// Timeout — per-request budget. 0 ⇒ webfetch.DefaultTimeout.
+	Timeout Duration `json:"timeout,omitempty"`
+
+	// MaxBytes — response size cap. 0 ⇒ webfetch.DefaultMaxBytes (2 MiB).
+	MaxBytes int64 `json:"max_bytes,omitempty"`
+
+	// MaxRunes — how much extracted text is kept. 0 ⇒ crawl.DefaultMaxRunes
+	// (4000), which is what fits a 4096-token context alongside a prompt.
+	MaxRunes int `json:"max_runes,omitempty"`
+}
+
+// CrawlWatchConfig — one page kept an eye on.
+type CrawlWatchConfig struct {
+	Name     string   `json:"name"` // note source is "crawl:<name>"
+	URL      string   `json:"url"`
+	Interval Duration `json:"interval,omitempty"` // 0 ⇒ CrawlConfig.Interval
+}
+
 // MemoryEvalConfig — the background memory-evaluation loop (Vikunja #248).
 // Absent ⇒ off, like every other capability that costs something the owner did
 // not ask for. Each evaluation is a full LLM round-trip on the one resident
@@ -519,6 +579,21 @@ type PhraserConfig struct {
 	// persona and invented units). Chat, query and reminder phrasing always go
 	// through the model regardless. See phraser.Config.LLMNudges.
 	LLMNudges bool `json:"llm_nudges,omitempty"`
+
+	// SwapModels — the gguf files the running daemon is allowed to swap to
+	// without a restart (Vikunja #250). Empty (the default) means the swap
+	// capability does not exist: ipc.MethodSwapModel answers ErrUnknownMethod,
+	// exactly like an unconfigured weather or telegram block.
+	//
+	// It is an allowlist and not a directory on purpose. The request carries a
+	// path, and llama-server is started with it as `-m`; anything short of an
+	// exact match against a list a human wrote in this file would make "swap the
+	// model" mean "load a file of your choosing off my disk". ModelPath is
+	// always swappable back to whether or not it is listed.
+	//
+	// Paths must be absolute — the daemon's working directory is not the
+	// operator's, and a relative path here would resolve somewhere surprising.
+	SwapModels []string `json:"swap_models,omitempty"`
 }
 
 // EmbedderConfig — paths for the ONNX multilingual embedder. The daemon
@@ -697,6 +772,12 @@ func (c *Config) applyDefaults() {
 		c.Feeds = nil
 	}
 
+	// Same rule for the crawler: a block that neither answers on demand nor
+	// watches anything has nothing to do, so it is normalised to "off".
+	if c.Crawl != nil && !c.Crawl.OnDemand && len(c.Crawl.Watches) == 0 {
+		c.Crawl = nil
+	}
+
 	if c.Voice != nil {
 		if c.Voice.RouterThreshold <= 0 {
 			c.Voice.RouterThreshold = DefaultRouterThreshold
@@ -754,6 +835,14 @@ func (c *Config) validate() error {
 		if c.Phraser.ModelPath == "" {
 			return errors.New("phraser.model_path is required")
 		}
+		// A relative entry in the swap allowlist would resolve against the
+		// daemon's working directory, so the path a human reads in this file
+		// would not be the path llama-server is handed. Fail at startup.
+		for _, m := range c.Phraser.SwapModels {
+			if !filepath.IsAbs(m) {
+				return fmt.Errorf("phraser.swap_models: %q must be an absolute path", m)
+			}
+		}
 	}
 	if c.Voice != nil && c.Voice.Enabled {
 		if c.Voice.Bind == "" {

internal/config/config_test.go

@@ -293,3 +293,35 @@ func TestPatternProposalNotifyDefaultsOff(t *testing.T) {
 		t.Errorf("cooldown = %v, want 6h", c.PatternProposals.Cooldown)
 	}
 }
+
+// TestSwapModelsAbsentMeansOff — the swap capability does not exist unless the
+// operator lists the models he allows (Vikunja #250).
+func TestSwapModelsAbsentMeansOff(t *testing.T) {
+	c, err := Load(writeConfig(t, `{"phraser": {"model_path": "/m/qwen.gguf"}}`))
+	if err != nil {
+		t.Fatalf("Load: %v", err)
+	}
+	if len(c.Phraser.SwapModels) != 0 {
+		t.Errorf("swap_models = %v; want empty when unconfigured", c.Phraser.SwapModels)
+	}
+}
+
+func TestSwapModelsParsedAndMustBeAbsolute(t *testing.T) {
+	c, err := Load(writeConfig(t, `{"phraser": {
+		"model_path": "/m/qwen.gguf",
+		"swap_models": ["/m/qwen.gguf", "/m/qwen-cpt.gguf"]
+	}}`))
+	if err != nil {
+		t.Fatalf("Load: %v", err)
+	}
+	if len(c.Phraser.SwapModels) != 2 {
+		t.Fatalf("swap_models = %v; want 2 entries", c.Phraser.SwapModels)
+	}
+	// A relative entry would resolve against the daemon's cwd, not the operator's.
+	if _, err := Load(writeConfig(t, `{"phraser": {
+		"model_path": "/m/qwen.gguf",
+		"swap_models": ["models/llm/qwen.gguf"]
+	}}`)); err == nil {
+		t.Error("Load accepted a relative swap_models entry; want a startup failure")
+	}
+}

internal/crawl/crawl.go

@@ -0,0 +1,174 @@
+// Package crawl reads a web page: fetch, robots check, HTML to text.
+//
+// It is the LAST place Maven looks for an answer, and that ordering is the whole
+// design. "Never phones home" is deprecated, but what replaced it puts local
+// sources first: the resident model, then his own memory, then the Kiwix ZIMs on
+// the box (internal/kiwix), and only then the network. A local read costs
+// nothing and leaks nothing; a fetch costs a round-trip and puts a URL in
+// someone's access log. So this package exists to be the fallback, not the
+// front door — see the querySources chain in cmd/mavend/actions_query.go for
+// where it actually sits.
+//
+// What never leaves the box: his notes, his facts, the persona block, the
+// conversation history. Only the URL is requested and, for the on-demand path,
+// only because he said it out loud. Nothing here reads the store.
+//
+// The limits are not in this package — they are in internal/webfetch, which is
+// the only way anything here touches a socket: http(s) only, host allow/deny,
+// private-address refusal, size cap, redirect cap, per-host rate limit. What
+// this package adds is politeness (robots.txt) and dedup.
+package crawl
+
+import (
+	"context"
+	"crypto/sha256"
+	"encoding/hex"
+	"errors"
+	"fmt"
+	"net/url"
+	"strings"
+	"time"
+)
+
+// Errors callers distinguish.
+var (
+	ErrRobots  = errors.New("crawl: robots.txt disallows this path")
+	ErrNotHTML = errors.New("crawl: response is not html or text")
+)
+
+// Fetcher is the guarded HTTP door (internal/webfetch adapted by the daemon). An
+// interface so this package constructs no http.Client of its own and can be
+// tested without a network.
+type Fetcher interface {
+	Get(ctx context.Context, url string) (*Response, error)
+}
+
+// Response is the minimum a crawl needs from a fetch.
+type Response struct {
+	URL         string
+	ContentType string
+	Body        []byte
+}
+
+// Config — crawler knobs.
+type Config struct {
+	// UserAgent is the name matched against robots.txt groups. It must be the
+	// same string the fetcher sends, or Maven would be claiming one identity
+	// and obeying the rules for another.
+	UserAgent string
+	// MaxRunes caps extracted text. 0 ⇒ DefaultMaxRunes.
+	MaxRunes int
+	// RobotsTTL — how long a parsed robots.txt is trusted. 0 ⇒ 1h.
+	RobotsTTL time.Duration
+	// Now is injectable for tests. nil ⇒ time.Now.
+	Now func() time.Time
+}
+
+// Crawler fetches and extracts pages. Safe for concurrent use.
+type Crawler struct {
+	fetch  Fetcher
+	cfg    Config
+	robots *robotsCache
+}
+
+// New builds a crawler. Returns nil when there is no fetcher, which is how the
+// daemon expresses "crawling is off unless configured".
+func New(fetch Fetcher, cfg Config) *Crawler {
+	if fetch == nil {
+		return nil
+	}
+	if cfg.UserAgent == "" {
+		cfg.UserAgent = "Maven"
+	}
+	if cfg.MaxRunes <= 0 {
+		cfg.MaxRunes = DefaultMaxRunes
+	}
+	if cfg.RobotsTTL <= 0 {
+		cfg.RobotsTTL = time.Hour
+	}
+	if cfg.Now == nil {
+		cfg.Now = time.Now
+	}
+	return &Crawler{fetch: fetch, cfg: cfg, robots: newRobotsCache(cfg.RobotsTTL)}
+}
+
+// Page fetches rawURL and returns its text. It checks robots.txt first and
+// refuses a disallowed path with ErrRobots — there is no override.
+func (c *Crawler) Page(ctx context.Context, rawURL string) (Page, error) {
+	u, err := url.Parse(strings.TrimSpace(rawURL))
+	if err != nil {
+		return Page{}, fmt.Errorf("crawl: bad url %q: %w", rawURL, err)
+	}
+	ok, err := c.allowed(ctx, u)
+	if err != nil {
+		return Page{}, err
+	}
+	if !ok {
+		return Page{}, fmt.Errorf("%w: %s", ErrRobots, u.Path)
+	}
+	resp, err := c.fetch.Get(ctx, u.String())
+	if err != nil {
+		return Page{}, err
+	}
+	// A PDF or an image is bytes Maven cannot read; saying so beats storing
+	// binary garbage as a "note".
+	ct := strings.ToLower(resp.ContentType)
+	if ct != "" && !strings.Contains(ct, "html") && !strings.Contains(ct, "text/") &&
+		!strings.Contains(ct, "xml") && !strings.Contains(ct, "json") {
+		return Page{}, fmt.Errorf("%w: %s", ErrNotHTML, resp.ContentType)
+	}
+	return Extract(resp.URL, resp.Body, c.cfg.MaxRunes), nil
+}
+
+// allowed consults robots.txt for u's host, reading it at most once per TTL.
+//
+// A robots.txt that cannot be fetched (404, a timeout, a blocked host) means
+// allow, per the standard. The one thing that is NOT fail-open is an explicit
+// Disallow.
+func (c *Crawler) allowed(ctx context.Context, u *url.URL) (bool, error) {
+	host := u.Host
+	now := c.cfg.Now()
+	rules, ok := c.robots.get(host, now)
+	if !ok {
+		robotsURL := u.Scheme + "://" + host + "/robots.txt"
+		resp, err := c.fetch.Get(ctx, robotsURL)
+		switch {
+		case err != nil:
+			// Note what is NOT swallowed: a refusal from the guarded fetcher.
+			// If webfetch says this host is denied or private, the page fetch
+			// would fail the same way, and reporting the real reason beats
+			// reporting a robots verdict we never got.
+			if isFatalFetchError(err) {
+				return false, err
+			}
+			rules = Rules{}
+		default:
+			rules = ParseRobots(string(resp.Body), c.cfg.UserAgent)
+		}
+		c.robots.put(host, rules, now)
+	}
+	path := u.EscapedPath()
+	if u.RawQuery != "" {
+		path += "?" + u.RawQuery
+	}
+	return rules.Allowed(path), nil
+}
+
+// isFatalFetchError — a fetch failure that means "this host is off limits"
+// rather than "there is no robots.txt here". The sentinel set is webfetch's, but
+// this package must not import it (the interface exists precisely so it does
+// not), so the check is on the message. Ugly and honest: the alternative is a
+// dependency inversion for two strings.
+func isFatalFetchError(err error) bool {
+	s := err.Error()
+	return strings.Contains(s, "not allowed") || strings.Contains(s, "private address") ||
+		strings.Contains(s, "only http and https")
+}
+
+// Hash is the dedup key for a crawl result: the sha256 of the extracted text,
+// hex, first 16 chars. Text and not raw HTML, because a page whose only change
+// is a rotating ad slot or a CSRF token has not changed.
+func Hash(text string) string {
+	sum := sha256.Sum256([]byte(strings.TrimSpace(text)))
+	return hex.EncodeToString(sum[:])[:16]
+}

internal/crawl/crawl_test.go

@@ -0,0 +1,155 @@
+package crawl
+
+import (
+	"context"
+	"errors"
+	"strings"
+	"testing"
+	"time"
+)
+
+// fakeFetcher serves canned pages by URL and counts requests, so a test can
+// assert that robots.txt was read once and that a refusal never reached the page.
+type fakeFetcher struct {
+	pages map[string]Response
+	err   error
+	calls []string
+}
+
+func (f *fakeFetcher) Get(_ context.Context, u string) (*Response, error) {
+	f.calls = append(f.calls, u)
+	if f.err != nil {
+		return nil, f.err
+	}
+	r, ok := f.pages[u]
+	if !ok {
+		return nil, errors.New("http 404")
+	}
+	if r.URL == "" {
+		r.URL = u
+	}
+	if r.ContentType == "" {
+		r.ContentType = "text/html; charset=utf-8"
+	}
+	return &r, nil
+}
+
+const htmlPage = `<html><head><title>Почему небо синее</title>
+<style>body{color:red}</style><script>track()</script></head>
+<body><nav>меню</nav><h1>Небо</h1>
+<p>Свет рассеивается на молекулах воздуха.</p>
+<p>Короткие волны рассеиваются сильнее.</p>
+<footer>© 2026</footer></body></html>`
+
+func newTestCrawler(f *fakeFetcher) *Crawler {
+	return New(f, Config{UserAgent: "Maven/1.0", Now: func() time.Time { return time.Unix(0, 0) }})
+}
+
+func TestPageExtractsText(t *testing.T) {
+	f := &fakeFetcher{pages: map[string]Response{
+		"https://example.org/sky": {Body: []byte(htmlPage)},
+	}}
+	page, err := newTestCrawler(f).Page(context.Background(), "https://example.org/sky")
+	if err != nil {
+		t.Fatal(err)
+	}
+	if page.Title != "Почему небо синее" {
+		t.Errorf("title = %q", page.Title)
+	}
+	if !strings.Contains(page.Text, "Свет рассеивается") {
+		t.Errorf("body text missing: %q", page.Text)
+	}
+	for _, junk := range []string{"track()", "color:red", "меню", "© 2026"} {
+		if strings.Contains(page.Text, junk) {
+			t.Errorf("%q survived extraction: %q", junk, page.Text)
+		}
+	}
+}
+
+func TestRobotsIsCheckedAndObeyed(t *testing.T) {
+	f := &fakeFetcher{pages: map[string]Response{
+		"https://example.org/robots.txt": {Body: []byte("User-agent: *\nDisallow: /secret\n"), ContentType: "text/plain"},
+		"https://example.org/secret/x":   {Body: []byte(htmlPage)},
+		"https://example.org/open":       {Body: []byte(htmlPage)},
+	}}
+	c := newTestCrawler(f)
+	if _, err := c.Page(context.Background(), "https://example.org/secret/x"); !errors.Is(err, ErrRobots) {
+		t.Fatalf("error = %v, want ErrRobots", err)
+	}
+	for _, u := range f.calls {
+		if strings.Contains(u, "/secret") {
+			t.Fatal("the disallowed page was fetched anyway")
+		}
+	}
+	if _, err := c.Page(context.Background(), "https://example.org/open"); err != nil {
+		t.Fatalf("allowed page: %v", err)
+	}
+	// robots.txt was read once for the host, not once per page.
+	robotsReads := 0
+	for _, u := range f.calls {
+		if strings.HasSuffix(u, "/robots.txt") {
+			robotsReads++
+		}
+	}
+	if robotsReads != 1 {
+		t.Fatalf("robots.txt read %d times, want 1", robotsReads)
+	}
+}
+
+// No robots.txt means allow — that is the standard, and the alternative makes
+// most of the web unreadable.
+func TestMissingRobotsAllows(t *testing.T) {
+	f := &fakeFetcher{pages: map[string]Response{
+		"https://example.org/page": {Body: []byte(htmlPage)},
+	}}
+	if _, err := newTestCrawler(f).Page(context.Background(), "https://example.org/page"); err != nil {
+		t.Fatalf("err = %v, want the page", err)
+	}
+}
+
+// A refusal from the guarded fetcher must surface as itself, not be laundered
+// into "no robots.txt, go ahead".
+func TestFetcherRefusalIsNotSwallowed(t *testing.T) {
+	f := &fakeFetcher{err: errors.New("webfetch: refusing to connect to a private address: 127.0.0.1")}
+	_, err := newTestCrawler(f).Page(context.Background(), "http://127.0.0.1:9100/mcp")
+	if err == nil || !strings.Contains(err.Error(), "private address") {
+		t.Fatalf("error = %v, want the fetcher's refusal", err)
+	}
+}
+
+func TestNonTextIsRefused(t *testing.T) {
+	f := &fakeFetcher{pages: map[string]Response{
+		"https://example.org/f.pdf": {Body: []byte("%PDF-1.7"), ContentType: "application/pdf"},
+	}}
+	if _, err := newTestCrawler(f).Page(context.Background(), "https://example.org/f.pdf"); !errors.Is(err, ErrNotHTML) {
+		t.Fatalf("error = %v, want ErrNotHTML", err)
+	}
+}
+
+func TestMaxRunesCapsText(t *testing.T) {
+	long := "<html><body><p>" + strings.Repeat("привет ", 2000) + "</p></body></html>"
+	f := &fakeFetcher{pages: map[string]Response{"https://example.org/l": {Body: []byte(long)}}}
+	c := New(f, Config{MaxRunes: 50})
+	page, err := c.Page(context.Background(), "https://example.org/l")
+	if err != nil {
+		t.Fatal(err)
+	}
+	if n := len([]rune(page.Text)); n > 51 {
+		t.Fatalf("text = %d runes, want the 50-rune cap", n)
+	}
+}
+
+func TestNewWithoutFetcherIsNil(t *testing.T) {
+	if New(nil, Config{}) != nil {
+		t.Fatal("a crawler with no fetcher must be nil — crawling is off unless configured")
+	}
+}
+
+func TestHashIgnoresNothingButText(t *testing.T) {
+	if Hash("a") == Hash("b") {
+		t.Fatal("different text hashed the same")
+	}
+	if Hash(" same \n") != Hash("same") {
+		t.Fatal("surrounding whitespace changed the hash")
+	}
+}

internal/crawl/extract.go

@@ -0,0 +1,106 @@
+package crawl
+
+import (
+	"html"
+	"regexp"
+	"strings"
+)
+
+// HTML → text, with a regexp and no tokenizer.
+//
+// golang.org/x/net/html is not vendored and the network is not assumed, so this
+// is stdlib. That is less of a compromise than it sounds: the unit of context
+// here is a few hundred words for a 4096-token model to read, exactly like the
+// Kiwix snippet, so what matters is dropping script/style/nav noise and keeping
+// paragraph boundaries. A DOM would buy correctness on malformed markup that is
+// then thrown away by truncation anyway.
+//
+// What this deliberately does NOT do: run JavaScript, follow links, or extract
+// structured fields with CSS selectors or an LLM prompt. The plan's step 2 asked
+// for the last of those; see docs/plans/14-web-crawler.md for why it was left
+// out for now.
+
+var (
+	// RE2 has no backreferences, so each tag pair is spelled out rather than
+	// captured and matched against itself.
+	dropRE  = regexp.MustCompile(pairsRE("script", "style", "noscript", "svg", "head", "nav", "footer", "form"))
+	titleRE = regexp.MustCompile(`(?is)<title\b[^>]*>(.*?)</title>`)
+	h1RE    = regexp.MustCompile(`(?is)<h1\b[^>]*>(.*?)</h1>`)
+	// Block-level tags become newlines so paragraphs survive as paragraphs.
+	blockRE   = regexp.MustCompile(`(?is)</?(p|div|br|li|tr|h[1-6]|section|article|blockquote|pre)\b[^>]*>`)
+	tagRE     = regexp.MustCompile(`(?s)<[^>]*>`)
+	commentRE = regexp.MustCompile(`(?s)<!--.*?-->`)
+	spaceRE   = regexp.MustCompile(`[ \t\f\v]+`)
+	blankRE   = regexp.MustCompile(`\n{2,}`)
+)
+
+// pairsRE builds `(?is)<tag …>…</tag>|…` for the given tags.
+func pairsRE(tags ...string) string {
+	parts := make([]string, 0, len(tags))
+	for _, t := range tags {
+		parts = append(parts, `<`+t+`\b[^>]*>.*?</`+t+`>`)
+	}
+	return `(?is)` + strings.Join(parts, "|")
+}
+
+// Page is an extracted page.
+type Page struct {
+	URL   string
+	Title string
+	Text  string // plain text, paragraphs separated by single newlines
+}
+
+// Extract turns a fetched HTML document into a Page. maxRunes caps the text (0 ⇒
+// DefaultMaxRunes); the cap is on runes, not bytes, because a Russian page cut
+// at a byte boundary ends in half a letter.
+func Extract(url string, body []byte, maxRunes int) Page {
+	if maxRunes <= 0 {
+		maxRunes = DefaultMaxRunes
+	}
+	s := string(body)
+	s = commentRE.ReplaceAllString(s, " ")
+
+	title := firstGroup(titleRE, s)
+	if title == "" {
+		title = firstGroup(h1RE, s)
+	}
+
+	s = dropRE.ReplaceAllString(s, "\n")
+	s = blockRE.ReplaceAllString(s, "\n")
+	s = tagRE.ReplaceAllString(s, " ")
+	s = html.UnescapeString(s)
+	s = spaceRE.ReplaceAllString(s, " ")
+
+	var lines []string
+	for _, l := range strings.Split(s, "\n") {
+		if l = strings.TrimSpace(l); l != "" {
+			lines = append(lines, l)
+		}
+	}
+	text := blankRE.ReplaceAllString(strings.Join(lines, "\n"), "\n")
+
+	return Page{URL: url, Title: title, Text: TrimRunes(text, maxRunes)}
+}
+
+// DefaultMaxRunes — how much of a page is kept. ~4000 runes is a long answer's
+// worth of context and still leaves room in a 4096-token window for the prompt
+// and the reply.
+const DefaultMaxRunes = 4000
+
+func firstGroup(re *regexp.Regexp, s string) string {
+	m := re.FindStringSubmatch(s)
+	if len(m) < 2 {
+		return ""
+	}
+	t := tagRE.ReplaceAllString(m[1], " ")
+	return strings.TrimSpace(strings.Join(strings.Fields(html.UnescapeString(t)), " "))
+}
+
+// TrimRunes cuts s to at most max runes, on a rune boundary.
+func TrimRunes(s string, max int) string {
+	r := []rune(s)
+	if len(r) <= max {
+		return s
+	}
+	return strings.TrimSpace(string(r[:max])) + "…"
+}

internal/crawl/robots.go

@@ -0,0 +1,211 @@
+package crawl
+
+import (
+	"regexp"
+	"strings"
+	"sync"
+	"time"
+)
+
+// robots.txt, parsed the small way: no wildcards beyond the two the standard
+// actually defines (`*` inside a path and `$` at the end), no sitemaps, no
+// crawl-delay-per-agent gymnastics. A personal assistant reading a handful of
+// pages does not need a spec-complete implementation; it needs to not be rude,
+// and to be auditable in one sitting.
+//
+// Two rules worth stating because they are choices, not accidents:
+//
+//   - a missing or unreadable robots.txt means ALLOW. That is what the standard
+//     says (404 ⇒ unrestricted), and the alternative would make a site that
+//     simply has no robots.txt unreadable;
+//   - an explicit Disallow means REFUSE, and Maven does not offer an override.
+//     There is no "but he asked me to" flag: the page is not read.
+
+// Rules is a parsed robots.txt for one user-agent.
+type Rules struct {
+	allow    []string
+	disallow []string
+	// Delay is Crawl-delay in seconds when the group named one, 0 otherwise.
+	// The fetcher's own per-host rate limit is the floor; this can only make
+	// Maven slower, never faster.
+	Delay time.Duration
+}
+
+// ParseRobots reads robots.txt and returns the rules that apply to agent.
+//
+// Group selection follows the standard: the most specific matching group wins,
+// which here means an exact user-agent match beats `*`. Lines that are neither
+// are ignored rather than guessed at.
+func ParseRobots(body string, agent string) Rules {
+	agent = strings.ToLower(agent)
+
+	type group struct {
+		agents   []string
+		allow    []string
+		disallow []string
+		delay    time.Duration
+	}
+	var groups []group
+	var cur *group
+	// startNew tracks whether the next User-agent line opens a new group or
+	// joins the current one: consecutive User-agent lines share their rules.
+	startNew := true
+
+	for _, raw := range strings.Split(body, "\n") {
+		line := raw
+		if i := strings.IndexByte(line, '#'); i >= 0 {
+			line = line[:i]
+		}
+		line = strings.TrimSpace(line)
+		if line == "" {
+			continue
+		}
+		key, val, ok := strings.Cut(line, ":")
+		if !ok {
+			continue
+		}
+		key = strings.ToLower(strings.TrimSpace(key))
+		val = strings.TrimSpace(val)
+
+		switch key {
+		case "user-agent":
+			if startNew || cur == nil {
+				groups = append(groups, group{})
+				cur = &groups[len(groups)-1]
+				startNew = false
+			}
+			cur.agents = append(cur.agents, strings.ToLower(val))
+		case "disallow":
+			if cur == nil {
+				continue
+			}
+			startNew = true
+			// "Disallow:" with an empty value allows everything, and is not a
+			// path rule at all.
+			if val != "" {
+				cur.disallow = append(cur.disallow, val)
+			}
+		case "allow":
+			if cur == nil {
+				continue
+			}
+			startNew = true
+			if val != "" {
+				cur.allow = append(cur.allow, val)
+			}
+		case "crawl-delay":
+			if cur == nil {
+				continue
+			}
+			startNew = true
+			if d, err := time.ParseDuration(val + "s"); err == nil && d > 0 {
+				cur.delay = d
+			}
+		}
+	}
+
+	var star, exact *group
+	for i := range groups {
+		for _, a := range groups[i].agents {
+			if a == "*" && star == nil {
+				star = &groups[i]
+			}
+			// A robots.txt names "maven", we send "Maven/1.0 (…)": match on
+			// prefix, which is how every crawler reads this field.
+			if a != "*" && a != "" && strings.HasPrefix(agent, a) {
+				exact = &groups[i]
+			}
+		}
+	}
+	g := exact
+	if g == nil {
+		g = star
+	}
+	if g == nil {
+		return Rules{}
+	}
+	return Rules{allow: g.allow, disallow: g.disallow, Delay: g.delay}
+}
+
+// Allowed reports whether path may be fetched. Longest matching rule wins, and
+// Allow beats Disallow at equal length — the standard's tie-break, and the one
+// that makes "Disallow: /" plus "Allow: /public" mean what it looks like.
+func (r Rules) Allowed(path string) bool {
+	if path == "" {
+		path = "/"
+	}
+	best, allowed := -1, true
+	for _, p := range r.disallow {
+		if n, ok := matchPath(p, path); ok && n > best {
+			best, allowed = n, false
+		}
+	}
+	for _, p := range r.allow {
+		if n, ok := matchPath(p, path); ok && n >= best {
+			best, allowed = n, true
+		}
+	}
+	return allowed
+}
+
+// matchPath applies a robots path pattern and returns the pattern's length as
+// the specificity score. `*` matches any run of characters, `$` anchors the end.
+// A pattern is a PREFIX match otherwise, which is what "Disallow: /admin" means.
+func matchPath(pattern, path string) (int, bool) {
+	score := len(pattern)
+	re, err := robotsRegexp(pattern)
+	if err != nil {
+		return 0, false
+	}
+	return score, re.MatchString(path)
+}
+
+// robotsRegexp turns a robots path pattern into an anchored-at-the-start
+// regexp. Everything but `*` and a trailing `$` is a literal, so the pattern is
+// quoted first and the two metacharacters are put back afterwards.
+func robotsRegexp(pattern string) (*regexp.Regexp, error) {
+	end := ""
+	if strings.HasSuffix(pattern, "$") {
+		pattern = strings.TrimSuffix(pattern, "$")
+		end = "$"
+	}
+	parts := strings.Split(pattern, "*")
+	for i, p := range parts {
+		parts[i] = regexp.QuoteMeta(p)
+	}
+	return regexp.Compile("^" + strings.Join(parts, ".*") + end)
+}
+
+// robotsCache holds parsed rules per host so a crawl of ten pages on one site
+// reads robots.txt once. TTL because a site may change its mind, and a daemon
+// that runs for weeks would otherwise never notice.
+type robotsCache struct {
+	ttl time.Duration
+	mu  sync.Mutex
+	m   map[string]robotsEntry
+}
+
+type robotsEntry struct {
+	rules Rules
+	at    time.Time
+}
+
+func newRobotsCache(ttl time.Duration) *robotsCache {
+	return &robotsCache{ttl: ttl, m: map[string]robotsEntry{}}
+}
+
+func (c *robotsCache) get(host string, now time.Time) (Rules, bool) {
+	c.mu.Lock()
+	defer c.mu.Unlock()
+	e, ok := c.m[host]
+	if !ok || now.Sub(e.at) > c.ttl {
+		return Rules{}, false
+	}
+	return e.rules, true
+}
+
+func (c *robotsCache) put(host string, r Rules, now time.Time) {
+	c.mu.Lock()
+	defer c.mu.Unlock()
+	c.m[host] = robotsEntry{rules: r, at: now}
+}

internal/crawl/robots_test.go

@@ -0,0 +1,84 @@
+package crawl
+
+import (
+	"testing"
+	"time"
+)
+
+const robotsBody = `# a comment
+User-agent: *
+Disallow: /private
+Disallow: /tmp/
+Crawl-delay: 5
+
+User-agent: Maven
+Disallow: /
+Allow: /public
+`
+
+func TestParseRobotsPicksTheMostSpecificGroup(t *testing.T) {
+	// The Maven group applies to us even though we send a longer UA string.
+	r := ParseRobots(robotsBody, "Maven/1.0 (self-hosted personal assistant)")
+	if r.Allowed("/anything") {
+		t.Error("Disallow: / in our own group was ignored")
+	}
+	if !r.Allowed("/public/page") {
+		t.Error("Allow: /public must beat the shorter Disallow: /")
+	}
+
+	// A different agent falls into the * group.
+	star := ParseRobots(robotsBody, "SomeoneElse/2")
+	if !star.Allowed("/anything") {
+		t.Error("the * group disallows nothing but /private and /tmp/")
+	}
+	if star.Allowed("/private/x") || star.Allowed("/tmp/") {
+		t.Error("the * group's disallows were not applied")
+	}
+	if star.Delay != 5*time.Second {
+		t.Errorf("crawl-delay = %v, want 5s", star.Delay)
+	}
+}
+
+func TestParseRobotsEmptyMeansAllowAll(t *testing.T) {
+	for _, body := range []string{"", "# nothing here\n", "User-agent: *\nDisallow:\n"} {
+		if !ParseRobots(body, "Maven").Allowed("/whatever") {
+			t.Errorf("body %q must allow everything", body)
+		}
+	}
+}
+
+func TestRobotsWildcards(t *testing.T) {
+	r := ParseRobots("User-agent: *\nDisallow: /*.pdf$\nDisallow: /a/*/secret\n", "Maven")
+	if r.Allowed("/docs/manual.pdf") {
+		t.Error("*.pdf$ did not match")
+	}
+	if !r.Allowed("/docs/manual.pdf.html") {
+		t.Error("$ must anchor at the end")
+	}
+	if r.Allowed("/a/b/secret") {
+		t.Error("/a/*/secret did not match")
+	}
+	if !r.Allowed("/a/b/public") {
+		t.Error("unrelated path was refused")
+	}
+}
+
+// Consecutive User-agent lines share one group, which is common in the wild.
+func TestRobotsSharedGroup(t *testing.T) {
+	r := ParseRobots("User-agent: Googlebot\nUser-agent: Maven\nDisallow: /x\n", "Maven/1.0")
+	if r.Allowed("/x/y") {
+		t.Fatal("a shared group's rules were not applied to the second agent")
+	}
+}
+
+func TestRobotsCacheTTL(t *testing.T) {
+	c := newRobotsCache(time.Minute)
+	now := time.Now()
+	c.put("example.com", ParseRobots("User-agent: *\nDisallow: /\n", "Maven"), now)
+	if _, ok := c.get("example.com", now.Add(30*time.Second)); !ok {
+		t.Error("a fresh entry must be served from cache")
+	}
+	if _, ok := c.get("example.com", now.Add(2*time.Minute)); ok {
+		t.Error("an expired entry must be re-read")
+	}
+}

internal/crawl/watch.go

@@ -0,0 +1,177 @@
+package crawl
+
+import (
+	"context"
+	"fmt"
+	"log"
+	"strings"
+	"time"
+)
+
+// Scheduled crawls: a page is re-read on an interval, and when its TEXT changed
+// the new text is written as a note. Nothing is dispatched — same rule as the
+// feed poller (Vikunja #258). A page that announced its own change would be a
+// nag, and "the docs page changed" is not worth interrupting anyone for.
+//
+// Dedup is by content hash, so a page that re-renders identically writes nothing
+// and a rotating ad slot does not count as news.
+
+// WatchConfig — one page to keep an eye on.
+type WatchConfig struct {
+	Name     string        // note source is "crawl:<Name>"
+	URL      string        // http(s), guarded by the fetcher
+	Interval time.Duration // 0 ⇒ Watcher's default
+}
+
+// Notes is core's note-writing half (same shape as ipc.CoreAPI's method).
+type Notes interface {
+	WriteNote(ctx context.Context, ts time.Time, text string, embedding []float32, source string) (int64, error)
+}
+
+// Hashes remembers the last text hash per watch, durably, so a restart does not
+// re-note an unchanged page. The daemon backs this with config facts
+// ("crawl:hash:<name>").
+type Hashes interface {
+	LastHash(ctx context.Context, name string) (string, error)
+	SetHash(ctx context.Context, name, hash string) error
+}
+
+// Embedder embeds a note on its way into the store. nil ⇒ no vector.
+type Embedder interface {
+	Embed(ctx context.Context, text string) ([]float32, error)
+}
+
+// DefaultWatchInterval — pages change slowly, and every check is a request in
+// someone's log.
+const DefaultWatchInterval = 6 * time.Hour
+
+// Watcher re-reads watched pages on their interval.
+type Watcher struct {
+	c        *Crawler
+	watches  []WatchConfig
+	notes    Notes
+	hashes   Hashes
+	embed    Embedder
+	interval time.Duration
+	nextDue  map[string]time.Time
+}
+
+// NewWatcher wires the scheduled half, or returns nil when there is nothing to
+// watch. Callers check for nil: no watches, no goroutine, no request.
+func NewWatcher(c *Crawler, watches []WatchConfig, notes Notes, hashes Hashes, embed Embedder, defaultInterval time.Duration) *Watcher {
+	if c == nil || notes == nil {
+		return nil
+	}
+	var valid []WatchConfig
+	for _, w := range watches {
+		if strings.TrimSpace(w.Name) == "" || strings.TrimSpace(w.URL) == "" {
+			log.Printf("crawl: skipping a watch with no name or no url")
+			continue
+		}
+		valid = append(valid, w)
+	}
+	if len(valid) == 0 {
+		return nil
+	}
+	if defaultInterval <= 0 {
+		defaultInterval = DefaultWatchInterval
+	}
+	return &Watcher{
+		c: c, watches: valid, notes: notes, hashes: hashes, embed: embed,
+		interval: defaultInterval, nextDue: map[string]time.Time{},
+	}
+}
+
+// Watches returns the configured watches.
+func (w *Watcher) Watches() []WatchConfig { return w.watches }
+
+// CheckDue re-reads every watch whose interval elapsed and returns how many
+// notes were written. Errors are logged per watch, never returned: one dead page
+// must not stop the others.
+func (w *Watcher) CheckDue(ctx context.Context, now time.Time) int {
+	written := 0
+	for _, watch := range w.watches {
+		if due, ok := w.nextDue[watch.Name]; ok && now.Before(due) {
+			continue
+		}
+		interval := watch.Interval
+		if interval <= 0 {
+			interval = w.interval
+		}
+		w.nextDue[watch.Name] = now.Add(interval)
+		changed, err := w.Check(ctx, watch, now)
+		if err != nil {
+			log.Printf("crawl: watch %s: %v", watch.Name, err)
+			continue
+		}
+		if changed {
+			log.Printf("crawl: watch %s: page changed, noted", watch.Name)
+			written++
+		}
+	}
+	return written
+}
+
+// Check re-reads one watch now and reports whether it wrote a note.
+func (w *Watcher) Check(ctx context.Context, watch WatchConfig, now time.Time) (bool, error) {
+	page, err := w.c.Page(ctx, watch.URL)
+	if err != nil {
+		return false, err
+	}
+	// Title included: a page whose headline changed has changed.
+	h := Hash(page.Title + "\n" + page.Text)
+	if w.hashes != nil {
+		prev, err := w.hashes.LastHash(ctx, watch.Name)
+		if err != nil {
+			log.Printf("crawl: watch %s: read hash: %v", watch.Name, err)
+		}
+		if prev == h {
+			return false, nil
+		}
+	}
+	text := NoteText(watch, page)
+	var vec []float32
+	if w.embed != nil {
+		v, err := w.embed.Embed(ctx, text)
+		if err != nil {
+			log.Printf("crawl: watch %s: embed: %v", watch.Name, err)
+		} else {
+			vec = v
+		}
+	}
+	if _, err := w.notes.WriteNote(ctx, now, text, vec, SourceFor(watch.Name)); err != nil {
+		return false, fmt.Errorf("write note: %w", err)
+	}
+	if w.hashes != nil {
+		if err := w.hashes.SetHash(ctx, watch.Name, h); err != nil {
+			log.Printf("crawl: watch %s: save hash: %v", watch.Name, err)
+		}
+	}
+	return true, nil
+}
+
+// SourceFor is the note source for a watch, and SourcePrefix is what the answer
+// path matches to recognise one.
+func SourceFor(name string) string { return SourcePrefix + name }
+
+// SourcePrefix — provenance for anything read off the network on a schedule.
+const SourcePrefix = "crawl:"
+
+// noteRunes — how much of a watched page goes into a note. Shorter than what the
+// on-demand path reads: a note is a record of a change, not an archive.
+const noteRunes = 800
+
+// NoteText renders a watched page as a note body.
+func NoteText(watch WatchConfig, page Page) string {
+	var b strings.Builder
+	if page.Title != "" {
+		b.WriteString(page.Title)
+	} else {
+		b.WriteString(watch.Name)
+	}
+	b.WriteString("\n")
+	b.WriteString(TrimRunes(page.Text, noteRunes))
+	b.WriteString("\n")
+	b.WriteString(watch.URL)
+	return b.String()
+}

internal/crawl/watch_test.go

@@ -0,0 +1,117 @@
+package crawl
+
+import (
+	"context"
+	"strings"
+	"testing"
+	"time"
+)
+
+type note struct {
+	text   string
+	source string
+}
+
+type fakeNotes struct{ notes []note }
+
+func (n *fakeNotes) WriteNote(_ context.Context, _ time.Time, text string, _ []float32, source string) (int64, error) {
+	n.notes = append(n.notes, note{text, source})
+	return int64(len(n.notes)), nil
+}
+
+type fakeHashes struct{ m map[string]string }
+
+func newHashes() *fakeHashes { return &fakeHashes{m: map[string]string{}} }
+func (f *fakeHashes) LastHash(_ context.Context, name string) (string, error) {
+	return f.m[name], nil
+}
+func (f *fakeHashes) SetHash(_ context.Context, name, h string) error { f.m[name] = h; return nil }
+
+var t0 = time.Date(2026, 8, 1, 9, 0, 0, 0, time.UTC)
+
+func TestWatchNotesAChangedPage(t *testing.T) {
+	f := &fakeFetcher{pages: map[string]Response{
+		"https://example.org/docs": {Body: []byte(htmlPage)},
+	}}
+	notes := &fakeNotes{}
+	hashes := newHashes()
+	w := NewWatcher(newTestCrawler(f), []WatchConfig{{Name: "docs", URL: "https://example.org/docs"}},
+		notes, hashes, nil, time.Hour)
+	if w == nil {
+		t.Fatal("NewWatcher returned nil for a configured watch")
+	}
+	if n := w.CheckDue(context.Background(), t0); n != 1 {
+		t.Fatalf("first check wrote %d notes, want 1", n)
+	}
+	if notes.notes[0].source != "crawl:docs" {
+		t.Errorf("source = %q, want crawl:docs", notes.notes[0].source)
+	}
+	if !strings.Contains(notes.notes[0].text, "https://example.org/docs") {
+		t.Errorf("note does not carry the url: %q", notes.notes[0].text)
+	}
+
+	// Unchanged page, interval elapsed: nothing written.
+	if n := w.CheckDue(context.Background(), t0.Add(2*time.Hour)); n != 0 {
+		t.Fatalf("an unchanged page wrote %d notes", n)
+	}
+
+	// Changed page: one note.
+	f.pages["https://example.org/docs"] = Response{Body: []byte(strings.Replace(htmlPage, "синее", "серое", 1))}
+	if n := w.CheckDue(context.Background(), t0.Add(4*time.Hour)); n != 1 {
+		t.Fatalf("a changed page wrote %d notes, want 1", n)
+	}
+}
+
+func TestWatchIntervalIsRespected(t *testing.T) {
+	f := &fakeFetcher{pages: map[string]Response{"https://example.org/d": {Body: []byte(htmlPage)}}}
+	w := NewWatcher(newTestCrawler(f), []WatchConfig{{Name: "d", URL: "https://example.org/d", Interval: time.Hour}},
+		&fakeNotes{}, newHashes(), nil, 0)
+	w.CheckDue(context.Background(), t0)
+	before := len(f.calls)
+	w.CheckDue(context.Background(), t0.Add(time.Minute))
+	if len(f.calls) != before {
+		t.Fatal("the page was re-read inside its interval")
+	}
+}
+
+// The hash is durable so a restart does not re-note an unchanged page.
+func TestWatchHashSurvivesRestart(t *testing.T) {
+	f := &fakeFetcher{pages: map[string]Response{"https://example.org/d": {Body: []byte(htmlPage)}}}
+	hashes := newHashes()
+	watches := []WatchConfig{{Name: "d", URL: "https://example.org/d"}}
+	NewWatcher(newTestCrawler(f), watches, &fakeNotes{}, hashes, nil, time.Hour).CheckDue(context.Background(), t0)
+
+	notes2 := &fakeNotes{}
+	NewWatcher(newTestCrawler(f), watches, notes2, hashes, nil, time.Hour).CheckDue(context.Background(), t0.Add(time.Hour))
+	if len(notes2.notes) != 0 {
+		t.Fatalf("a fresh watcher re-noted an unchanged page: %q", notes2.notes[0].text)
+	}
+}
+
+func TestWatchDeadPageDoesNotStopTheOthers(t *testing.T) {
+	f := &fakeFetcher{pages: map[string]Response{"https://example.org/live": {Body: []byte(htmlPage)}}}
+	notes := &fakeNotes{}
+	w := NewWatcher(newTestCrawler(f), []WatchConfig{
+		{Name: "dead", URL: "https://example.org/gone"},
+		{Name: "live", URL: "https://example.org/live"},
+	}, notes, newHashes(), nil, time.Hour)
+	if n := w.CheckDue(context.Background(), t0); n != 1 {
+		t.Fatalf("wrote %d notes, want 1 (the live page)", n)
+	}
+	if notes.notes[0].source != "crawl:live" {
+		t.Fatalf("source = %q", notes.notes[0].source)
+	}
+}
+
+func TestNoWatchesMeansNoWatcher(t *testing.T) {
+	c := newTestCrawler(&fakeFetcher{})
+	if NewWatcher(c, nil, &fakeNotes{}, nil, nil, 0) != nil {
+		t.Fatal("no watches must mean no watcher")
+	}
+	if NewWatcher(nil, []WatchConfig{{Name: "a", URL: "u"}}, &fakeNotes{}, nil, nil, 0) != nil {
+		t.Fatal("no crawler must mean no watcher")
+	}
+	if NewWatcher(c, []WatchConfig{{Name: "", URL: ""}}, &fakeNotes{}, nil, nil, 0) != nil {
+		t.Fatal("a watch with no name or url is not a configuration")
+	}
+}

internal/ipc/api.go

@@ -176,6 +176,50 @@ type IngestMailResp struct {
 	Skipped bool    `json:"skipped,omitempty"`
 }
 
+// SwapModelReq — load another resident model without restarting the daemon
+// (Vikunja #250). ModelPath must be one of the paths in phraser.swap_models;
+// anything else is ErrForbidden, and an unconfigured allowlist makes the whole
+// method ErrUnknownMethod.
+//
+// NGpuLayers and NCtx are zero for "keep what is loaded now", which is the
+// normal case — the same laptop iGPU, a different gguf.
+//
+// This is an owner action. It is AuthStepUp in the authority table, it is not on
+// CoreAPI, and no act, intent or timer can reach it: swapping the model is not
+// something Maven does to herself.
+type SwapModelReq struct {
+	ModelPath  string `json:"model_path"`
+	NGpuLayers int    `json:"n_gpu_layers,omitempty"`
+	NCtx       int    `json:"n_ctx,omitempty"`
+}
+
+// SwapModelResp — what the daemon ended up serving. Model is the identity the
+// new llama-server reported for itself, not an echo of the request: if the file
+// was not the model the operator thought it was, this is where it shows.
+//
+// RolledBack is true when the requested model failed to load or would not answer
+// and the previous one was put back. In that case the call also returns an error
+// — the swap did not happen — and Model names the model still serving.
+type SwapModelResp struct {
+	Model      string `json:"model"`
+	ModelPath  string `json:"model_path"`
+	BaseURL    string `json:"base_url"`
+	RolledBack bool   `json:"rolled_back,omitempty"`
+	TookMs     int64  `json:"took_ms"`
+}
+
+// ModelStatusResp — which model is resident and which ones may be swapped in.
+// Read-only; the authed page renders it. Swappable is the configured allowlist,
+// so an empty list means the capability is off.
+type ModelStatusResp struct {
+	Model      string   `json:"model"`
+	ModelPath  string   `json:"model_path"`
+	BaseURL    string   `json:"base_url"`
+	NGpuLayers int      `json:"n_gpu_layers"`
+	NCtx       int      `json:"n_ctx"`
+	Swappable  []string `json:"swappable,omitempty"`
+}
+
 type listTasksReq struct {
 	Status string `json:"status"` // "" all | "live" | candidate|open|done|dropped
 }

internal/ipc/client.go

@@ -459,6 +459,28 @@ func (c *Client) IngestMail(ctx context.Context, req IngestMailReq) (IngestMailR
 	return r, nil
 }
 
+// SwapModel asks core to load another resident model (Vikunja #250).
+// ErrUnknownMethod means core has no phraser.swap_models allowlist configured;
+// ErrForbidden means the path is not on it, or step-up was not asserted. A
+// non-nil error with RolledBack set means nothing changed — the old model is
+// still serving.
+func (c *Client) SwapModel(ctx context.Context, req SwapModelReq) (SwapModelResp, error) {
+	var r SwapModelResp
+	if err := c.call(ctx, MethodSwapModel, req, &r); err != nil {
+		return SwapModelResp{}, err
+	}
+	return r, nil
+}
+
+// ModelStatus reports the resident model and the swap allowlist. Read-only.
+func (c *Client) ModelStatus(ctx context.Context) (ModelStatusResp, error) {
+	var r ModelStatusResp
+	if err := c.call(ctx, MethodModelStatus, nil, &r); err != nil {
+		return ModelStatusResp{}, err
+	}
+	return r, nil
+}
+
 func (c *Client) DismissProposedRoutine(ctx context.Context, id int64) error {
 	return c.call(ctx, MethodDismissProposedRoutine, dismissProposedRoutineReq{ID: id}, nil)
 }

internal/ipc/ipc_test.go

@@ -6,6 +6,7 @@ import (
 	"encoding/binary"
 	"encoding/json"
 	"errors"
+	"fmt"
 	"io"
 	"net"
 	"os"
@@ -598,3 +599,44 @@ func TestIngestMail_Hook(t *testing.T) {
 		t.Errorf("req across the wire = %+v", got)
 	}
 }
+
+// TestSwapModel_OffUnlessConfigured — no allowlist in the config means the
+// daemon never sets the hook, and the method does not exist. That is what "off
+// unless configured" looks like at the wire for the model swap (Vikunja #250).
+func TestSwapModel_OffUnlessConfigured(t *testing.T) {
+	_, _, cli, _ := newServerWithStore(t)
+	if _, err := cli.SwapModel(context.Background(), SwapModelReq{ModelPath: "/m/x.gguf"}); !errors.Is(err, ErrUnknownMethod) {
+		t.Fatalf("SwapModel error = %v, want ErrUnknownMethod", err)
+	}
+	if _, err := cli.ModelStatus(context.Background()); !errors.Is(err, ErrUnknownMethod) {
+		t.Fatalf("ModelStatus error = %v, want ErrUnknownMethod", err)
+	}
+}
+
+// TestSwapModel_Hook — the request crosses the boundary intact and the reported
+// identity comes back. A refusal from the daemon's allowlist arrives as
+// ErrForbidden, which is what a caller keys its error message off.
+func TestSwapModel_Hook(t *testing.T) {
+	_, srv, cli, _ := newServerWithStore(t)
+	var got SwapModelReq
+	srv.SwapModelFn = func(_ context.Context, req SwapModelReq) (SwapModelResp, error) {
+		got = req
+		if req.ModelPath != "/m/allowed.gguf" {
+			return SwapModelResp{}, fmt.Errorf("%w: not allowlisted", ErrForbidden)
+		}
+		return SwapModelResp{Model: "allowed", ModelPath: req.ModelPath, BaseURL: "http://127.0.0.1:9", TookMs: 12}, nil
+	}
+	resp, err := cli.SwapModel(context.Background(), SwapModelReq{ModelPath: "/m/allowed.gguf", NCtx: 4096})
+	if err != nil {
+		t.Fatalf("SwapModel: %v", err)
+	}
+	if resp.Model != "allowed" || resp.TookMs != 12 {
+		t.Errorf("resp = %+v", resp)
+	}
+	if got.NCtx != 4096 {
+		t.Errorf("req across the wire = %+v", got)
+	}
+	if _, err := cli.SwapModel(context.Background(), SwapModelReq{ModelPath: "/etc/shadow"}); !errors.Is(err, ErrForbidden) {
+		t.Fatalf("swap to a non-allowlisted path = %v; want ErrForbidden", err)
+	}
+}

internal/ipc/server.go

@@ -432,6 +432,19 @@ type Server struct {
 	// every CoreAPI implementation has to carry.
 	IngestMailFn IngestMailFunc
 
+	// SwapModelFn / ModelStatusFn — the on-the-fly resident model swap (Vikunja
+	// #250) and its read side. Set by the daemon only when phraser.swap_models
+	// lists at least one model AND the phraser owns a llama-server; nil ⇒ both
+	// methods answer ErrUnknownMethod, which is what "off unless configured"
+	// looks like at the wire.
+	//
+	// They bypass CoreAPI for the same reason IngestMailFn does: this is not a
+	// store operation, it needs the daemon's llama-server, and no other CoreAPI
+	// implementation should have to carry it. MethodSwapModel is AuthStepUp in
+	// internal/auth — owner-triggered, never an act and never a timer.
+	SwapModelFn   SwapModelFunc
+	ModelStatusFn ModelStatusFunc
+
 	// UnlockFn — unwraps the store encryption key from the wrapped blob using
 	// the passkey credential public key, opens the encrypted store, and wires
 	// the rest of the daemon (voice, loop, delivery). Set by the daemon when
@@ -450,6 +463,12 @@ type WrapKeyFunc func(ctx context.Context, publicKey []byte) error
 // public key and completes daemon initialization.
 type UnlockFunc func(ctx context.Context, publicKey []byte) error
 
+// SwapModelFunc — loads another resident model in place of the live one.
+type SwapModelFunc func(ctx context.Context, req SwapModelReq) (SwapModelResp, error)
+
+// ModelStatusFunc — reports the resident model and the swap allowlist.
+type ModelStatusFunc func(ctx context.Context) (ModelStatusResp, error)
+
 // IngestMailFunc — core-side mail extraction. Returns what was captured.
 type IngestMailFunc func(ctx context.Context, req IngestMailReq) (IngestMailResp, error)
 
@@ -620,8 +639,9 @@ func withoutParams[R any](fn func(ctx context.Context, api CoreAPI) (R, error))
 // existed) as an argument — so SetAPI's runtime swap (the unlock transition)
 // is still honored on the very next request with no extra plumbing here.
 //
-// MethodAssertStepUp, MethodStoreEncryptionKey, MethodUnlock and
-// MethodIngestMail are NOT in this table: they bypass CoreAPI entirely
+// MethodAssertStepUp, MethodStoreEncryptionKey, MethodUnlock,
+// MethodIngestMail, MethodSwapModel and MethodModelStatus are NOT in this
+// table: they bypass CoreAPI entirely
 // (s.StepUp / s.WrapKeyFn / s.UnlockFn / s.IngestMailFn), so dispatch
 // special-cases them before consulting the table.
 var methodTable = map[Method]handlerFunc{
@@ -874,6 +894,30 @@ func (s *Server) dispatch(ctx context.Context, req Request) (json.RawMessage, er
 			return marshalResult(resp), nil
 		}
 		return nil, fmt.Errorf("%w: %s", ErrUnknownMethod, req.Method)
+
+	case MethodSwapModel:
+		if s.SwapModelFn != nil {
+			var p SwapModelReq
+			if err := unmarshalParams(req.Params, &p); err != nil {
+				return nil, err
+			}
+			resp, err := s.SwapModelFn(ctx, p)
+			if err != nil {
+				return nil, err
+			}
+			return marshalResult(resp), nil
+		}
+		return nil, fmt.Errorf("%w: %s", ErrUnknownMethod, req.Method)
+
+	case MethodModelStatus:
+		if s.ModelStatusFn != nil {
+			resp, err := s.ModelStatusFn(ctx)
+			if err != nil {
+				return nil, err
+			}
+			return marshalResult(resp), nil
+		}
+		return nil, fmt.Errorf("%w: %s", ErrUnknownMethod, req.Method)
 	}
 
 	h, ok := methodTable[req.Method]

internal/ipc/wire.go

@@ -51,6 +51,8 @@ const (
 	MethodListTasks              Method = "list_tasks"
 	MethodSetTaskStatus          Method = "set_task_status"
 	MethodIngestMail             Method = "ingest_mail"
+	MethodSwapModel              Method = "swap_model"
+	MethodModelStatus            Method = "model_status"
 )
 
 // Request — one frame from module to core. Params is the JSON-encoded argument

internal/llm/client.go

@@ -10,10 +10,17 @@ import (
 	"encoding/json"
 	"fmt"
 	"net/http"
+	"sync"
 	"time"
 )
 
 type Client struct {
+	// mu guards base only. The base URL changes when the daemon swaps the
+	// resident model (Vikunja #250): llama-server is relaunched on a fresh
+	// port, and every holder of this client — the LLM router, the replier, the
+	// mail extractor — must follow without being rebuilt. One mutexed field is
+	// the whole mechanism; a swap re-points the client, it does not replace it.
+	mu   sync.RWMutex
 	base string
 	http *http.Client
 }
@@ -22,6 +29,25 @@ func New(baseURL string, timeout time.Duration) *Client {
 	return &Client{base: baseURL, http: &http.Client{Timeout: timeout}}
 }
 
+// SetBaseURL re-points the client at another llama-server. Safe to call while
+// requests are in flight: a request that already read the old base finishes
+// against the old base (or fails, and every caller of Complete has a fallback),
+// and the next one uses the new base. It is deliberately NOT a queue-and-retry —
+// the phraser quiesces around a swap, so the window is small and a lost turn
+// degrades to the classifier rather than hanging.
+func (c *Client) SetBaseURL(base string) {
+	c.mu.Lock()
+	c.base = base
+	c.mu.Unlock()
+}
+
+// BaseURL is the server this client currently talks to.
+func (c *Client) BaseURL() string {
+	c.mu.RLock()
+	defer c.mu.RUnlock()
+	return c.base
+}
+
 type Req struct {
 	System    string
 	User      string
@@ -63,7 +89,7 @@ func (c *Client) Complete(ctx context.Context, r Req) (string, error) {
 		RepeatPenalty: r.RepeatPenalty,
 		Stop:          r.Stop,
 	})
-	req, err := http.NewRequestWithContext(ctx, "POST", c.base+"/v1/chat/completions", bytes.NewReader(b))
+	req, err := http.NewRequestWithContext(ctx, "POST", c.BaseURL()+"/v1/chat/completions", bytes.NewReader(b))
 	if err != nil {
 		return "", err
 	}

internal/llm/client_test.go

@@ -57,3 +57,37 @@ func TestComplete(t *testing.T) {
 		t.Errorf("got %q, want %q", got, "ok")
 	}
 }
+
+// TestSetBaseURL — a model swap re-points every holder of the client rather than
+// rebuilding the router, the replier and the extractors (Vikunja #250).
+func TestSetBaseURL(t *testing.T) {
+	var hit string
+	srvA := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		hit = "A"
+		w.Write([]byte(`{"choices":[{"message":{"content":"a"}}]}`))
+	}))
+	defer srvA.Close()
+	srvB := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		hit = "B"
+		w.Write([]byte(`{"choices":[{"message":{"content":"b"}}]}`))
+	}))
+	defer srvB.Close()
+
+	c := New(srvA.URL, 5*time.Second)
+	if _, err := c.Complete(context.Background(), Req{User: "x"}); err != nil {
+		t.Fatalf("Complete against A: %v", err)
+	}
+	if hit != "A" {
+		t.Fatalf("first request went to %q; want A", hit)
+	}
+	c.SetBaseURL(srvB.URL)
+	if got := c.BaseURL(); got != srvB.URL {
+		t.Errorf("BaseURL = %q; want %q", got, srvB.URL)
+	}
+	if _, err := c.Complete(context.Background(), Req{User: "x"}); err != nil {
+		t.Fatalf("Complete against B: %v", err)
+	}
+	if hit != "B" {
+		t.Errorf("request after the swap went to %q; want B", hit)
+	}
+}

internal/phraser/llmphraser.go

@@ -27,14 +27,46 @@ var listenRE = regexp.MustCompile(`listening on (https?://\S+)`)
 type LLMPhraser struct {
 	cfg    Config
 	client *http.Client
-	port   string
-	cmd    *exec.Cmd
-	cancel context.CancelFunc
-	wg     sync.WaitGroup
 
 	// tmpl — the hand-written Russian nudges. Default path for nudges; see
 	// Config.LLMNudges. nil only if the template file failed to load.
 	tmpl *NudgeTemplates
+
+	// spawnCtx — the parent of every llama-server this phraser starts, i.e. the
+	// daemon's own context. Deliberately NOT the per-request context of the call
+	// that asked for a model swap: that one is cancelled the moment the request
+	// returns, which would kill the model it had just loaded.
+	spawnCtx context.Context
+	cancel   context.CancelFunc
+
+	// launch / probe — the two side effects of a swap, injectable so the swap
+	// logic is testable without a real llama-server and a real model file.
+	// launch is nil when this phraser does not own its server (NewLLMPhraserAt),
+	// which is also what makes Swap refuse there.
+	launch func(ctx context.Context, cfg Config) (backend, error)
+	probe  func(ctx context.Context, base string) (string, error)
+
+	// swapMu — single-flight around Swap. Held for the whole swap, including the
+	// model load, so two concurrent swap requests can never both be loading.
+	swapMu sync.Mutex
+
+	// mu guards everything below: the live backend, the swap gate and the
+	// in-flight request count. See acquire/quiesce in swap.go.
+	mu        sync.Mutex
+	be        backend
+	live      liveModel
+	swapping  bool
+	inflight  int
+	observers []func(baseURL string)
+}
+
+// liveModel — what is actually loaded right now. Distinct from Config, which
+// stays immutable after construction: a swap changes these three fields and
+// nothing else, so no reader of cfg (prompts, grammar, timeouts) races a swap.
+type liveModel struct {
+	ModelPath  string
+	NGpuLayers int
+	NCtx       int
 }
 
 type Config struct {
@@ -85,15 +117,21 @@ func DefaultConfig(modelPath string) Config {
 func NewLLMPhraser(ctx context.Context, cfg Config) (*LLMPhraser, error) {
 	ctx, cancel := context.WithCancel(ctx)
 	p := &LLMPhraser{
-		cfg:    cfg,
-		client: &http.Client{Timeout: cfg.Timeout},
-		cancel: cancel,
-		tmpl:   loadNudgeTemplates(),
+		cfg:      cfg,
+		client:   &http.Client{Timeout: cfg.Timeout},
+		tmpl:     loadNudgeTemplates(),
+		spawnCtx: ctx,
+		cancel:   cancel,
+		launch:   spawnLlamaServer,
+		probe:    defaultProbe,
+		live:     liveModel{ModelPath: cfg.ModelPath, NGpuLayers: cfg.NGpuLayers, NCtx: cfg.NCtx},
 	}
-	if err := p.start(ctx); err != nil {
+	be, err := p.launch(ctx, cfg)
+	if err != nil {
 		cancel()
 		return nil, err
 	}
+	p.be = be
 	return p, nil
 }
 
@@ -106,11 +144,17 @@ func NewLLMPhraser(ctx context.Context, cfg Config) (*LLMPhraser, error) {
 // still uses NewLLMPhraser and still owns its own child process.
 func NewLLMPhraserAt(baseURL string, cfg Config) *LLMPhraser {
 	return &LLMPhraser{
-		cfg:    cfg,
-		client: &http.Client{Timeout: cfg.Timeout},
-		port:   strings.TrimSuffix(baseURL, "/"),
-		cancel: func() {},
-		tmpl:   loadNudgeTemplates(),
+		cfg:      cfg,
+		client:   &http.Client{Timeout: cfg.Timeout},
+		tmpl:     loadNudgeTemplates(),
+		spawnCtx: context.Background(),
+		cancel:   func() {},
+		probe:    defaultProbe,
+		// launch stays nil: we did not start this server, so we must not stop it.
+		// Swap therefore refuses here (ErrSwapNotOwned) instead of killing a
+		// server another process depends on.
+		be:   borrowedBackend(strings.TrimSuffix(baseURL, "/")),
+		live: liveModel{ModelPath: cfg.ModelPath, NGpuLayers: cfg.NGpuLayers, NCtx: cfg.NCtx},
 	}
 }
 
@@ -126,16 +170,66 @@ func loadNudgeTemplates() *NudgeTemplates {
 	return nt
 }
 
-func (p *LLMPhraser) start(ctx context.Context) error {
+// backend — one llama-server this phraser talks to. Two implementations: a
+// llamaProc we spawned and must reap, and a borrowedBackend someone else owns.
+type backend interface {
+	BaseURL() string
+	Close() error
+}
+
+// borrowedBackend — a server started and owned by someone else (the phrasing
+// scorer's shared llama-server). Closing it is a no-op by construction.
+type borrowedBackend string
+
+func (b borrowedBackend) BaseURL() string { return string(b) }
+func (b borrowedBackend) Close() error    { return nil }
+
+// llamaProc — a llama-server child process plus the goroutine reading its
+// stderr. Close kills and reaps it; see the Pdeathsig note in spawnLlamaServer.
+type llamaProc struct {
+	base   string
+	cmd    *exec.Cmd
+	cancel context.CancelFunc
+	wg     sync.WaitGroup
+}
+
+func (l *llamaProc) BaseURL() string { return l.base }
+
+func (l *llamaProc) Close() error {
+	l.cancel()
+	if l.cmd != nil && l.cmd.Process != nil {
+		_ = l.cmd.Process.Kill()
+		_ = l.cmd.Wait() // reap the process — without Wait, the child becomes a zombie
+	}
+	l.wg.Wait()
+	return nil
+}
+
+// spawnLlamaServer starts one llama-server for cfg and waits until it says which
+// address it is listening on. ctx owns the process lifetime, so it must be the
+// daemon's context, not a request's.
+func spawnLlamaServer(ctx context.Context, cfg Config) (backend, error) {
+	ctx, cancel := context.WithCancel(ctx)
+	p, err := startLlamaProc(ctx, cfg)
+	if err != nil {
+		cancel()
+		return nil, err
+	}
+	p.cancel = cancel
+	return p, nil
+}
+
+func startLlamaProc(ctx context.Context, cfg Config) (*llamaProc, error) {
+	p := &llamaProc{}
 	args := []string{
-		"-m", p.cfg.ModelPath,
+		"-m", cfg.ModelPath,
 		"--host", "127.0.0.1",
-		"--port", extractPort(p.cfg.Listen),
-		"-c", fmt.Sprintf("%d", p.cfg.NCtx),
-		"-ngl", fmt.Sprintf("%d", p.cfg.NGpuLayers),
+		"--port", extractPort(cfg.Listen),
+		"-c", fmt.Sprintf("%d", cfg.NCtx),
+		"-ngl", fmt.Sprintf("%d", cfg.NGpuLayers),
 		"--no-webui",
 	}
-	cmd := exec.CommandContext(ctx, p.cfg.BinPath, args...)
+	cmd := exec.CommandContext(ctx, cfg.BinPath, args...)
 	// Pdeathsig: the kernel SIGKILLs llama-server the moment mavend dies — by
 	// ANY means, including SIGKILL/OOM/panic where our Close() never runs. Without
 	// it a hard-killed mavend orphans its llama-server (reparented to init, keeps
@@ -148,12 +242,12 @@ func (p *LLMPhraser) start(ctx context.Context) error {
 
 	stderr, err := cmd.StderrPipe()
 	if err != nil {
-		return fmt.Errorf("llm: stderr pipe: %w", err)
+		return nil, fmt.Errorf("llm: stderr pipe: %w", err)
 	}
 
 	if err := cmd.Start(); err != nil {
 		stderr.Close()
-		return fmt.Errorf("llm: start: %w", err)
+		return nil, fmt.Errorf("llm: start: %w", err)
 	}
 
 	portCh := make(chan string, 1)
@@ -186,32 +280,44 @@ func (p *LLMPhraser) start(ctx context.Context) error {
 
 	select {
 	case addr := <-portCh:
-		p.port = addr
-		return nil
+		p.base = addr
+		return p, nil
 	case err := <-errCh:
 		_ = cmd.Process.Kill()
 		_ = cmd.Wait()
-		return fmt.Errorf("llm: server output: %w", err)
+		return nil, fmt.Errorf("llm: server output: %w", err)
 	case <-ctx.Done():
 		_ = cmd.Process.Kill()
 		_ = cmd.Wait()
-		return ctx.Err()
+		return nil, ctx.Err()
 	case <-time.After(60 * time.Second):
 		_ = cmd.Process.Kill()
 		_ = cmd.Wait()
-		return fmt.Errorf("llm: server did not start within 60s")
+		return nil, fmt.Errorf("llm: server did not start within 60s")
 	}
 }
 
-func (p *LLMPhraser) BaseURL() string { return p.port }
+// BaseURL is the llama-server this phraser talks to right now. It changes when
+// the model is swapped, so callers that cache it must register an observer
+// (OnSwap) rather than keeping the string forever.
+func (p *LLMPhraser) BaseURL() string {
+	p.mu.Lock()
+	defer p.mu.Unlock()
+	if p.be == nil {
+		return ""
+	}
+	return p.be.BaseURL()
+}
 
 func (p *LLMPhraser) Close() error {
 	p.cancel()
-	if p.cmd != nil && p.cmd.Process != nil {
-		_ = p.cmd.Process.Kill()
-		_ = p.cmd.Wait() // reap the process — without Wait, the child becomes a zombie
+	p.mu.Lock()
+	be := p.be
+	p.be = nil
+	p.mu.Unlock()
+	if be != nil {
+		return be.Close()
 	}
-	p.wg.Wait()
 	return nil
 }
 
@@ -350,6 +456,11 @@ func chatSystemPrompt(block func() string) string {
 // the LLM completion endpoint. Like chatWithSystem but for an arbitrary message
 // slice — the caller owns the system prompt placement.
 func (p *LLMPhraser) chatWithMessages(ctx context.Context, msgs []chatMsg, maxTokens int) (string, error) {
+	base, release, err := p.acquire()
+	if err != nil {
+		return "", err
+	}
+	defer release()
 	req := chatReq{
 		Messages:    msgs,
 		Temperature: 0.7,
@@ -360,7 +471,7 @@ func (p *LLMPhraser) chatWithMessages(ctx context.Context, msgs []chatMsg, maxTo
 	if err != nil {
 		return "", fmt.Errorf("llm: marshal: %w", err)
 	}
-	httpReq, err := http.NewRequestWithContext(ctx, "POST", p.port+"/v1/chat/completions", bytes.NewReader(body))
+	httpReq, err := http.NewRequestWithContext(ctx, "POST", base+"/v1/chat/completions", bytes.NewReader(body))
 	if err != nil {
 		return "", fmt.Errorf("llm: request: %w", err)
 	}
@@ -493,6 +604,11 @@ func (p *LLMPhraser) chat(ctx context.Context, userPrompt string) (string, error
 }
 
 func (p *LLMPhraser) chatWithSystem(ctx context.Context, system, user string, maxTokens int) (string, error) {
+	base, release, err := p.acquire()
+	if err != nil {
+		return "", err
+	}
+	defer release()
 	req := chatReq{
 		Messages: []chatMsg{
 			{Role: "system", Content: system},
@@ -507,7 +623,7 @@ func (p *LLMPhraser) chatWithSystem(ctx context.Context, system, user string, ma
 		return "", fmt.Errorf("llm: marshal: %w", err)
 	}
 
-	httpReq, err := http.NewRequestWithContext(ctx, "POST", p.port+"/v1/chat/completions", bytes.NewReader(body))
+	httpReq, err := http.NewRequestWithContext(ctx, "POST", base+"/v1/chat/completions", bytes.NewReader(body))
 	if err != nil {
 		return "", fmt.Errorf("llm: request: %w", err)
 	}

internal/phraser/swap.go

@@ -0,0 +1,330 @@
+package phraser
+
+import (
+	"context"
+	"errors"
+	"fmt"
+	"log"
+	"time"
+
+	"github.com/kami/maven/internal/llm"
+)
+
+// Swapping the resident model without restarting the daemon (Vikunja #250).
+//
+// Three properties this file exists to hold, in order of importance:
+//
+//  1. NEVER two models resident at once. The deploy target is a laptop iGPU
+//     with the whole 1.7B offloaded to it (`n_gpu_layers: 99`); loading a second
+//     model beside the first is how you OOM the box, and a blue/green swap that
+//     "keeps the old one warm until the new one answers" does exactly that. So
+//     the old server is killed FIRST and the new one loaded after. The cost of
+//     that ordering is a window with no model at all, which is why:
+//
+//  2. A swap is atomic from a turn's point of view. An in-flight turn finishes
+//     on the old model — Swap waits for the last one to return before killing
+//     anything. A turn that arrives during the swap is REFUSED immediately with
+//     ErrSwapping rather than blocked: every phrasing path already has a
+//     fallback (templates, "вот что я нашла", the classifier for routing), so a
+//     fast refusal degrades one turn instead of hanging it for the length of a
+//     model load. No turn ever gets half of one model and half of another.
+//
+//  3. A failed load rolls back to the model that was working. The new server is
+//     probed (it must say which model it loaded) before it is published; if the
+//     launch or the probe fails, the previous config is relaunched and the
+//     phraser goes back to serving. Only if the rollback ALSO fails is the
+//     phraser left without a backend, and then it says so loudly and every turn
+//     degrades rather than breaks.
+//
+// Not here, deliberately: nothing calls Swap on a timer, and no act or intent can
+// reach it. It is an IPC method behind the step-up gate, i.e. owner-triggered.
+
+var (
+	// ErrSwapping — a turn arrived while the model was being swapped. Callers
+	// treat it like any other LLM error and use their fallback.
+	ErrSwapping = errors.New("phraser: model swap in progress")
+
+	// ErrSwapNotOwned — this phraser did not start its llama-server, so it must
+	// not stop one (NewLLMPhraserAt: the eval harness shares a server).
+	ErrSwapNotOwned = errors.New("phraser: llama-server is not ours to swap")
+
+	// ErrNoBackend — no model is loaded at all. Only reachable after a failed
+	// swap whose rollback also failed.
+	ErrNoBackend = errors.New("phraser: no llama-server loaded")
+
+	// ErrSwapBusy — a turn was still running when the drain deadline expired, so
+	// the swap was abandoned. Nothing was killed; ask again.
+	ErrSwapBusy = errors.New("phraser: turns still in flight, swap abandoned")
+)
+
+// SwapSpec — what to load. Zero NGpuLayers/NCtx keep whatever is live, so the
+// common case ("same settings, different gguf") is one field.
+type SwapSpec struct {
+	ModelPath  string
+	NGpuLayers int
+	NCtx       int
+}
+
+// SwapResult — what happened. Model is the identity the NEW server reported, so
+// it is evidence rather than an echo of the request: if the file at ModelPath is
+// not what the operator thought it was, this is where that shows up.
+type SwapResult struct {
+	Model      string
+	BaseURL    string
+	ModelPath  string
+	RolledBack bool
+	Took       time.Duration
+}
+
+// drainTimeout — how long Swap waits for in-flight turns before giving up. A
+// turn is at most Config.Timeout (30s in deploy) plus the model's own latency;
+// 90s covers a slow Thinking generation without wedging the caller forever.
+const drainTimeout = 90 * time.Second
+
+// probeTimeout — how long the new server gets to answer "which model do you
+// have". The load itself is bounded by spawnLlamaServer's own 60s wait.
+const probeTimeout = 30 * time.Second
+
+// defaultProbe asks the server which model it has loaded. This is the health
+// check: a server that answers /v1/models has finished loading weights and is
+// serving, and its answer is the identity we report back.
+func defaultProbe(ctx context.Context, base string) (string, error) {
+	return llm.ModelID(ctx, base)
+}
+
+// OnSwap registers a callback fired with the new base URL every time the live
+// backend changes, including after a rollback. Holders of an *llm.Client (the
+// LLM router, the replier, the mail extractor) register SetBaseURL here so a
+// swap re-points them without rebuilding the router or the handler.
+//
+// Callbacks run with no lock held, in registration order.
+func (p *LLMPhraser) OnSwap(fn func(baseURL string)) {
+	if fn == nil {
+		return
+	}
+	p.mu.Lock()
+	p.observers = append(p.observers, fn)
+	p.mu.Unlock()
+}
+
+// LiveModel is the model file currently loaded (and its load settings). Empty
+// ModelPath means no model is loaded.
+func (p *LLMPhraser) LiveModel() (path string, nGpuLayers, nCtx int) {
+	p.mu.Lock()
+	defer p.mu.Unlock()
+	return p.live.ModelPath, p.live.NGpuLayers, p.live.NCtx
+}
+
+// acquire reserves a slot for one request and returns the base URL to use.
+// Every request path must call it and must call the returned release exactly
+// once — that count is what Swap drains.
+func (p *LLMPhraser) acquire() (string, func(), error) {
+	p.mu.Lock()
+	defer p.mu.Unlock()
+	if p.swapping {
+		return "", nil, ErrSwapping
+	}
+	if p.be == nil {
+		return "", nil, ErrNoBackend
+	}
+	p.inflight++
+	base := p.be.BaseURL()
+	var once bool
+	return base, func() {
+		p.mu.Lock()
+		if !once {
+			once = true
+			p.inflight--
+		}
+		p.mu.Unlock()
+	}, nil
+}
+
+// Swap loads another model in place of the live one. See the file comment for
+// the properties it guarantees. Returns the new model's reported identity, or
+// an error plus RolledBack=true when the old model was put back.
+//
+// ctx bounds the drain and the probe. It does NOT own the new server's lifetime
+// — that is the daemon's context, captured at construction — so a swap survives
+// the request that asked for it.
+func (p *LLMPhraser) Swap(ctx context.Context, spec SwapSpec) (SwapResult, error) {
+	if spec.ModelPath == "" {
+		return SwapResult{}, fmt.Errorf("phraser: swap needs a model path")
+	}
+	p.swapMu.Lock()
+	defer p.swapMu.Unlock()
+
+	if p.launch == nil {
+		return SwapResult{}, ErrSwapNotOwned
+	}
+
+	started := time.Now()
+	oldLive := p.liveSnapshot()
+	newLive := liveModel{
+		ModelPath:  spec.ModelPath,
+		NGpuLayers: pickInt(spec.NGpuLayers, oldLive.NGpuLayers),
+		NCtx:       pickInt(spec.NCtx, oldLive.NCtx),
+	}
+	if newLive == oldLive && p.BaseURL() != "" {
+		// Already serving exactly this. Report the live identity rather than
+		// pointlessly unloading and reloading the same weights.
+		base := p.BaseURL()
+		id, err := p.probeWith(ctx, base)
+		if err != nil {
+			return SwapResult{}, err
+		}
+		return SwapResult{Model: id, BaseURL: base, ModelPath: oldLive.ModelPath, Took: time.Since(started)}, nil
+	}
+
+	if err := p.quiesce(ctx); err != nil {
+		return SwapResult{}, err
+	}
+	defer p.resume()
+
+	// Property 1: the old model leaves the GPU before the new one arrives.
+	p.mu.Lock()
+	old := p.be
+	p.be = nil
+	p.mu.Unlock()
+	if old != nil {
+		_ = old.Close()
+	}
+
+	be, err := p.loadAndProbe(ctx, newLive)
+	if err != nil {
+		log.Printf("phraser: swap to %s FAILED (%v) — rolling back to %s", newLive.ModelPath, err, oldLive.ModelPath)
+		rb, rbErr := p.loadAndProbe(ctx, oldLive)
+		if rbErr != nil {
+			log.Printf("phraser: ROLLBACK to %s ALSO FAILED (%v) — no model is loaded, every phrasing path is on its fallback and routing is on the classifier until the daemon is restarted", oldLive.ModelPath, rbErr)
+			return SwapResult{RolledBack: true, Took: time.Since(started)},
+				fmt.Errorf("phraser: swap failed (%w) and rollback failed too: %v", err, rbErr)
+		}
+		p.publish(rb, oldLive)
+		return SwapResult{
+				Model: rb.id, BaseURL: rb.be.BaseURL(), ModelPath: oldLive.ModelPath,
+				RolledBack: true, Took: time.Since(started),
+			},
+			fmt.Errorf("phraser: swap to %s failed, rolled back to %s: %w", newLive.ModelPath, oldLive.ModelPath, err)
+	}
+	p.publish(be, newLive)
+	log.Printf("phraser: model swapped to %s (%s) at %s in %s", newLive.ModelPath, be.id, be.be.BaseURL(), time.Since(started).Round(time.Millisecond))
+	return SwapResult{
+		Model: be.id, BaseURL: be.be.BaseURL(), ModelPath: newLive.ModelPath,
+		Took: time.Since(started),
+	}, nil
+}
+
+// loaded — a started server plus the identity it reported.
+type loaded struct {
+	be backend
+	id string
+}
+
+// loadAndProbe starts a server for lm and verifies it answers. A server that
+// starts but will not say what it loaded is treated as a failed load and is
+// killed here — publishing it would hand every turn to a backend we could not
+// confirm.
+func (p *LLMPhraser) loadAndProbe(ctx context.Context, lm liveModel) (loaded, error) {
+	cfg := p.cfg
+	cfg.ModelPath = lm.ModelPath
+	cfg.NGpuLayers = lm.NGpuLayers
+	cfg.NCtx = lm.NCtx
+	// p.spawnCtx, not ctx: the process must outlive the request asking for it.
+	be, err := p.launch(p.spawnCtx, cfg)
+	if err != nil {
+		return loaded{}, err
+	}
+	id, err := p.probeWith(ctx, be.BaseURL())
+	if err != nil {
+		_ = be.Close()
+		return loaded{}, fmt.Errorf("phraser: %s started but would not answer: %w", lm.ModelPath, err)
+	}
+	return loaded{be: be, id: id}, nil
+}
+
+func (p *LLMPhraser) probeWith(ctx context.Context, base string) (string, error) {
+	probe := p.probe
+	if probe == nil {
+		probe = defaultProbe
+	}
+	pctx, cancel := context.WithTimeout(ctx, probeTimeout)
+	defer cancel()
+	return probe(pctx, base)
+}
+
+// quiesce closes the door on new turns and waits for the ones already running.
+// Polling rather than a sync.Cond: the wait happens once per swap, a 25ms poll
+// is invisible next to a model load, and a poll cannot deadlock on a release
+// path that panicked.
+func (p *LLMPhraser) quiesce(ctx context.Context) error {
+	p.mu.Lock()
+	if p.swapping {
+		p.mu.Unlock()
+		return ErrSwapping
+	}
+	p.swapping = true
+	inflight := p.inflight
+	p.mu.Unlock()
+	if inflight == 0 {
+		return nil
+	}
+
+	deadline := time.Now().Add(drainTimeout)
+	for {
+		select {
+		case <-ctx.Done():
+			p.resume()
+			return ctx.Err()
+		case <-time.After(25 * time.Millisecond):
+		}
+		p.mu.Lock()
+		inflight = p.inflight
+		p.mu.Unlock()
+		if inflight == 0 {
+			return nil
+		}
+		if time.Now().After(deadline) {
+			// Nothing has been killed yet, so abandoning is free: reopen the door
+			// and let the operator try again rather than cutting a live turn off
+			// mid-generation.
+			p.resume()
+			return fmt.Errorf("%w (%d still running after %s)", ErrSwapBusy, inflight, drainTimeout)
+		}
+	}
+}
+
+func (p *LLMPhraser) resume() {
+	p.mu.Lock()
+	p.swapping = false
+	p.mu.Unlock()
+}
+
+// publish makes l the live backend and tells everyone holding a base URL.
+func (p *LLMPhraser) publish(l loaded, lm liveModel) {
+	p.mu.Lock()
+	p.be = l.be
+	p.live = lm
+	obs := make([]func(string), len(p.observers))
+	copy(obs, p.observers)
+	p.mu.Unlock()
+	base := l.be.BaseURL()
+	for _, fn := range obs {
+		fn(base)
+	}
+}
+
+func (p *LLMPhraser) liveSnapshot() liveModel {
+	p.mu.Lock()
+	defer p.mu.Unlock()
+	return p.live
+}
+
+// pickInt returns v when the caller set it, and fallback otherwise. 0 is the
+// "unset" value: -1 already means "offload every layer" and deploy uses 99, so
+// nothing legitimate asks for exactly zero GPU layers through this path.
+func pickInt(v, fallback int) int {
+	if v == 0 {
+		return fallback
+	}
+	return v
+}

internal/phraser/swap_test.go

@@ -0,0 +1,311 @@
+package phraser
+
+import (
+	"context"
+	"errors"
+	"net/http"
+	"net/http/httptest"
+	"sync"
+	"sync/atomic"
+	"testing"
+	"time"
+)
+
+// fakeModel — a stand-in llama-server. It answers /v1/models with its own name
+// and /v1/chat/completions with a phrasing-contract reply that names itself, so
+// a test can tell WHICH model answered a turn — the property the swap is about.
+type fakeModel struct {
+	srv    *httptest.Server
+	name   string
+	closed atomic.Bool
+}
+
+func newFakeModel(t *testing.T, name string) *fakeModel {
+	t.Helper()
+	f := &fakeModel{name: name}
+	mux := http.NewServeMux()
+	mux.HandleFunc("/v1/models", func(w http.ResponseWriter, r *http.Request) {
+		w.Write([]byte(`{"data":[{"id":"/models/` + name + `.gguf"}]}`))
+	})
+	mux.HandleFunc("/v1/chat/completions", func(w http.ResponseWriter, r *http.Request) {
+		w.Write([]byte(`{"choices":[{"message":{"content":"{\"response\":\"` + name + `\",\"mood\":\"neutral\"}"}}]}`))
+	})
+	f.srv = httptest.NewServer(mux)
+	t.Cleanup(f.srv.Close)
+	return f
+}
+
+func (f *fakeModel) BaseURL() string { return f.srv.URL }
+func (f *fakeModel) Close() error    { f.closed.Store(true); return nil }
+
+// fakeFleet is the injected launcher: it hands out a prepared fakeModel per
+// model path, and refuses paths the test did not prepare (that is what a bad
+// gguf looks like from here). It also asserts the invariant that matters on a
+// laptop iGPU: never two servers alive at the same time.
+type fakeFleet struct {
+	mu      sync.Mutex
+	models  map[string]string // model path → fake name
+	live    int
+	maxLive int
+	launch  int
+}
+
+func (fl *fakeFleet) launcher(t *testing.T) func(context.Context, Config) (backend, error) {
+	return func(ctx context.Context, cfg Config) (backend, error) {
+		fl.mu.Lock()
+		name, ok := fl.models[cfg.ModelPath]
+		fl.launch++
+		if !ok {
+			fl.mu.Unlock()
+			return nil, errors.New("no such model file: " + cfg.ModelPath)
+		}
+		fl.live++
+		if fl.live > fl.maxLive {
+			fl.maxLive = fl.live
+		}
+		fl.mu.Unlock()
+		f := newFakeModel(t, name)
+		return &fleetBackend{fleet: fl, model: f}, nil
+	}
+}
+
+type fleetBackend struct {
+	fleet *fakeFleet
+	model *fakeModel
+	once  sync.Once
+}
+
+func (b *fleetBackend) BaseURL() string { return b.model.BaseURL() }
+func (b *fleetBackend) Close() error {
+	b.once.Do(func() {
+		b.fleet.mu.Lock()
+		b.fleet.live--
+		b.fleet.mu.Unlock()
+	})
+	return b.model.Close()
+}
+
+// newSwapPhraser builds an LLMPhraser with an injected launcher, so the swap
+// path is exercised without a gguf or a GPU.
+func newSwapPhraser(t *testing.T, fl *fakeFleet, modelPath string) *LLMPhraser {
+	t.Helper()
+	cfg := DefaultConfig(modelPath)
+	cfg.Timeout = 5 * time.Second
+	p := &LLMPhraser{
+		cfg:      cfg,
+		client:   &http.Client{Timeout: cfg.Timeout},
+		spawnCtx: context.Background(),
+		cancel:   func() {},
+		launch:   fl.launcher(t),
+		probe:    defaultProbe,
+		live:     liveModel{ModelPath: modelPath, NGpuLayers: cfg.NGpuLayers, NCtx: cfg.NCtx},
+	}
+	be, err := p.launch(p.spawnCtx, cfg)
+	if err != nil {
+		t.Fatalf("initial launch: %v", err)
+	}
+	p.be = be
+	t.Cleanup(func() { p.Close() })
+	return p
+}
+
+func TestSwap_LoadsNewModelAndRepointsHolders(t *testing.T) {
+	fl := &fakeFleet{models: map[string]string{"/m/old.gguf": "old", "/m/new.gguf": "new"}}
+	p := newSwapPhraser(t, fl, "/m/old.gguf")
+
+	// A holder of the base URL (the LLM router's client, in the daemon).
+	var seen []string
+	p.OnSwap(func(base string) { seen = append(seen, base) })
+
+	before, err := p.PhraseChat(context.Background(), "привет", nil)
+	if err != nil || before != "old" {
+		t.Fatalf("before swap: %q, %v; want the old model to answer", before, err)
+	}
+
+	res, err := p.Swap(context.Background(), SwapSpec{ModelPath: "/m/new.gguf"})
+	if err != nil {
+		t.Fatalf("Swap: %v", err)
+	}
+	if res.Model != "new" {
+		t.Errorf("res.Model = %q; want the identity the NEW server reported (%q)", res.Model, "new")
+	}
+	if res.RolledBack {
+		t.Errorf("res.RolledBack = true on a successful swap")
+	}
+	after, err := p.PhraseChat(context.Background(), "привет", nil)
+	if err != nil || after != "new" {
+		t.Fatalf("after swap: %q, %v; want the new model to answer", after, err)
+	}
+	if path, _, _ := p.LiveModel(); path != "/m/new.gguf" {
+		t.Errorf("LiveModel = %q; want /m/new.gguf", path)
+	}
+	if len(seen) != 1 || seen[0] != p.BaseURL() {
+		t.Errorf("observers saw %v; want exactly one call with the new base %q", seen, p.BaseURL())
+	}
+	if fl.maxLive > 1 {
+		t.Errorf("%d servers were alive at once; the iGPU only fits one model", fl.maxLive)
+	}
+}
+
+func TestSwap_FailedLoadRollsBackToTheWorkingModel(t *testing.T) {
+	fl := &fakeFleet{models: map[string]string{"/m/old.gguf": "old"}}
+	p := newSwapPhraser(t, fl, "/m/old.gguf")
+
+	res, err := p.Swap(context.Background(), SwapSpec{ModelPath: "/m/broken.gguf"})
+	if err == nil {
+		t.Fatal("Swap to a model that will not load returned nil error")
+	}
+	if !res.RolledBack {
+		t.Errorf("res.RolledBack = false; a failed swap must say it rolled back")
+	}
+	if res.Model != "old" {
+		t.Errorf("res.Model = %q; want the old model back", res.Model)
+	}
+	// The point of the rollback: turns keep working.
+	got, err := p.PhraseChat(context.Background(), "привет", nil)
+	if err != nil || got != "old" {
+		t.Fatalf("after rollback: %q, %v; want the old model serving again", got, err)
+	}
+	if path, _, _ := p.LiveModel(); path != "/m/old.gguf" {
+		t.Errorf("LiveModel = %q; want the old model", path)
+	}
+	if fl.maxLive > 1 {
+		t.Errorf("%d servers alive at once during a rollback", fl.maxLive)
+	}
+}
+
+func TestSwap_ProbeFailureIsTreatedAsAFailedLoad(t *testing.T) {
+	// A server that starts but will not say what it loaded must never be
+	// published — we would be serving turns from a backend we cannot confirm.
+	fl := &fakeFleet{models: map[string]string{"/m/old.gguf": "old", "/m/mute.gguf": "mute"}}
+	p := newSwapPhraser(t, fl, "/m/old.gguf")
+	// Fail the probe once — for the newly launched server — and let the
+	// rollback's probe through.
+	calls := 0
+	p.probe = func(ctx context.Context, base string) (string, error) {
+		calls++
+		if calls == 1 {
+			return "", errors.New("no answer from the new server")
+		}
+		return defaultProbe(ctx, base)
+	}
+
+	_, err := p.Swap(context.Background(), SwapSpec{ModelPath: "/m/mute.gguf"})
+	if err == nil {
+		t.Fatal("Swap published a server that failed its probe")
+	}
+	if path, _, _ := p.LiveModel(); path != "/m/old.gguf" {
+		t.Errorf("LiveModel = %q; want the old model after a failed probe", path)
+	}
+}
+
+func TestSwap_RollbackFailureLeavesNoBackendAndDegrades(t *testing.T) {
+	fl := &fakeFleet{models: map[string]string{"/m/old.gguf": "old"}}
+	p := newSwapPhraser(t, fl, "/m/old.gguf")
+	// Make the rollback fail too: the old file "disappears" mid-swap.
+	fl.mu.Lock()
+	delete(fl.models, "/m/old.gguf")
+	fl.mu.Unlock()
+
+	_, err := p.Swap(context.Background(), SwapSpec{ModelPath: "/m/broken.gguf"})
+	if err == nil {
+		t.Fatal("Swap returned nil when both the load and the rollback failed")
+	}
+	// Nothing is loaded, and the request path says so rather than panicking.
+	if _, _, aerr := p.acquire(); !errors.Is(aerr, ErrNoBackend) {
+		t.Errorf("acquire error = %v; want ErrNoBackend", aerr)
+	}
+	// Phrasing degrades to its fallback instead of failing the turn.
+	got, err := p.PhraseChat(context.Background(), "привет", nil)
+	if err != nil {
+		t.Fatalf("PhraseChat after a total failure returned an error: %v", err)
+	}
+	if got == "" {
+		t.Error("PhraseChat returned empty; the fallback must still say something")
+	}
+}
+
+func TestSwap_WaitsForInFlightTurnAndRefusesNewOnes(t *testing.T) {
+	fl := &fakeFleet{models: map[string]string{"/m/old.gguf": "old", "/m/new.gguf": "new"}}
+	p := newSwapPhraser(t, fl, "/m/old.gguf")
+
+	// Hold one turn open by taking a slot directly — the same slot every
+	// request path takes.
+	base, release, err := p.acquire()
+	if err != nil {
+		t.Fatalf("acquire: %v", err)
+	}
+	if base == "" {
+		t.Fatal("acquire returned an empty base URL")
+	}
+
+	swapped := make(chan error, 1)
+	go func() { _, e := p.Swap(context.Background(), SwapSpec{ModelPath: "/m/new.gguf"}); swapped <- e }()
+
+	// While the swap waits to drain, a NEW turn is refused immediately rather
+	// than blocked for the length of a model load.
+	deadline := time.Now().Add(2 * time.Second)
+	for {
+		_, rel, aerr := p.acquire()
+		if rel != nil {
+			rel()
+		}
+		if errors.Is(aerr, ErrSwapping) {
+			break
+		}
+		if time.Now().After(deadline) {
+			t.Fatalf("new turns were never refused during a swap (last error: %v)", aerr)
+		}
+		time.Sleep(10 * time.Millisecond)
+	}
+
+	// The swap cannot have completed while our turn was still in flight.
+	select {
+	case e := <-swapped:
+		t.Fatalf("Swap finished before the in-flight turn released: %v", e)
+	case <-time.After(50 * time.Millisecond):
+	}
+
+	release()
+	if e := <-swapped; e != nil {
+		t.Fatalf("Swap after drain: %v", e)
+	}
+	got, err := p.PhraseChat(context.Background(), "привет", nil)
+	if err != nil || got != "new" {
+		t.Fatalf("after swap: %q, %v; want the new model", got, err)
+	}
+}
+
+func TestSwap_RefusedWhenWeDoNotOwnTheServer(t *testing.T) {
+	// NewLLMPhraserAt points at a shared server the eval harness owns. Swapping
+	// there would kill a server another process depends on.
+	p := NewLLMPhraserAt("http://127.0.0.1:1/", DefaultConfig("/m/old.gguf"))
+	if _, err := p.Swap(context.Background(), SwapSpec{ModelPath: "/m/new.gguf"}); !errors.Is(err, ErrSwapNotOwned) {
+		t.Fatalf("Swap on a borrowed server = %v; want ErrSwapNotOwned", err)
+	}
+}
+
+func TestSwap_SameModelIsANoOp(t *testing.T) {
+	fl := &fakeFleet{models: map[string]string{"/m/old.gguf": "old"}}
+	p := newSwapPhraser(t, fl, "/m/old.gguf")
+	launchesBefore := fl.launch
+
+	res, err := p.Swap(context.Background(), SwapSpec{ModelPath: "/m/old.gguf"})
+	if err != nil {
+		t.Fatalf("Swap to the live model: %v", err)
+	}
+	if res.Model != "old" {
+		t.Errorf("res.Model = %q; want old", res.Model)
+	}
+	if fl.launch != launchesBefore {
+		t.Errorf("%d extra launches; swapping to the live model must not reload weights", fl.launch-launchesBefore)
+	}
+}
+
+func TestSwap_EmptyModelPathRefused(t *testing.T) {
+	fl := &fakeFleet{models: map[string]string{"/m/old.gguf": "old"}}
+	p := newSwapPhraser(t, fl, "/m/old.gguf")
+	if _, err := p.Swap(context.Background(), SwapSpec{}); err == nil {
+		t.Fatal("Swap with no model path returned nil error")
+	}
+}

internal/router/url.go

@@ -0,0 +1,39 @@
+package router
+
+import (
+	"regexp"
+	"strings"
+)
+
+// Finding a URL in an utterance (Vikunja #259).
+//
+// This is deliberately strict: a scheme is required. "посмотри на example.org"
+// is not treated as a fetch request, because a bare dotted word is also how
+// people say file names, versions and Russian abbreviations, and the cost of a
+// false positive here is an outbound request nobody asked for.
+//
+// Note where this runs: an utterance from STT. Whisper will mangle a spoken URL,
+// which is fine — the URL that survives is one he pasted into the web chat, and
+// a mangled one simply fails to match.
+var urlRE = regexp.MustCompile(`(?i)\bhttps?://[^\s<>"']+`)
+
+// FirstURL returns the first http(s) URL in text.
+//
+// Trailing punctuation is trimmed: he ends sentences, and "…/page." is not a
+// path component. A closing bracket is only trimmed when it has no opener,
+// because a wikipedia URL legitimately ends in one.
+func FirstURL(text string) (string, bool) {
+	m := urlRE.FindString(text)
+	if m == "" {
+		return "", false
+	}
+	m = strings.TrimRight(m, ".,;:!?…")
+	if strings.HasSuffix(m, ")") && strings.Count(m, "(") == 0 {
+		m = strings.TrimSuffix(m, ")")
+	}
+	// A scheme with nothing after it is not a URL.
+	if rest := strings.SplitN(m, "//", 2); len(rest) < 2 || rest[1] == "" {
+		return "", false
+	}
+	return m, true
+}

internal/router/url_test.go

@@ -0,0 +1,34 @@
+package router
+
+import "testing"
+
+func TestFirstURL(t *testing.T) {
+	cases := []struct {
+		text string
+		want string
+	}{
+		{"посмотри https://example.org/page — что там?", "https://example.org/page"},
+		{"почитай http://example.org/a/b?x=1 и скажи", "http://example.org/a/b?x=1"},
+		{"вот ссылка: https://example.org/page.", "https://example.org/page"},
+		{"https://ru.wikipedia.org/wiki/Небо_(значения)", "https://ru.wikipedia.org/wiki/Небо_(значения)"},
+		// No scheme ⇒ no fetch. A bare dotted word is not an instruction to
+		// reach out to the network.
+		{"посмотри на example.org", ""},
+		{"открой файл config.json", ""},
+		{"что нового?", ""},
+		{"https://", ""},
+		{"", ""},
+	}
+	for _, c := range cases {
+		got, ok := FirstURL(c.text)
+		if c.want == "" {
+			if ok {
+				t.Errorf("FirstURL(%q) = %q, want no match", c.text, got)
+			}
+			continue
+		}
+		if !ok || got != c.want {
+			t.Errorf("FirstURL(%q) = %q, %v; want %q", c.text, got, ok, c.want)
+		}
+	}
+}