fix(auth): rate-limit login per IP, 12-character passwords, loopback default

- POST /api/auth/login: 10 attempts per minute per TCP peer IP
  (X-Forwarded-For ignored), then 429 with Retry-After. The block starts
  at 1 minute and doubles on each attempt while blocked and on each new
  block, capped at 15 minutes. A successful login clears the IP. No
  per-account lockout. In-memory, idle entries swept after 15 minutes,
  at most 10,000 IPs.
- New passwords (init, change-password, /setup page) need 12 characters.
  Login does not check length, so existing 8-character passwords work.
- The binary's -addr default is 127.0.0.1:8787. The Docker image passes
  -addr :8787 explicitly and is unchanged.
- GET /api/health returns {"ok":true} without a valid token; with one it
  still returns service and version for the web UI top bar.

CHANGELOG.md

@@ -4,6 +4,11 @@ All notable changes to this project are documented in this file. The format is b
 
 ## [Unreleased]
 
+### Changed
+
+- **The binary listens on `127.0.0.1:8787` by default**: the `-addr` default was `:8787`, every interface. Running `artex` or `./start.sh` without `-addr` now accepts connections from the local machine only; pass `-addr :8787` to listen on every interface. Docker is unchanged: the image starts the server with `-addr :8787` inside the container, and `docker-compose.yml` already publishes it on `127.0.0.1` by default (`ARTEX_BIND_ADDR`).
+- **New passwords need at least 12 characters** (was 8), at first-run setup (`POST /api/auth/init` and the `/setup` page) and in `POST /api/auth/change-password`. Login does not check length, so a password set under the old minimum keeps working until it is changed. `reset-password.sh` does not check length.
+
 ### Fixed
 
 - **The JWT signing key survives container rebuilds**: the server wrote `jwt.key` to the executable's directory, `/app` in the image, so the key lived in the container's writable layer and recreating the container (for example `docker compose up -d --build`) generated a new key and signed every user out. The key directory is now set by `ARTEX_JWT_KEY_DIR` (default: the executable's directory, unchanged for non-Docker installs). `docker-compose.yml` sets it to `/app/keys` and bind-mounts `./keys` there, outside `./data`, which the file manager browses. On start, when `<key dir>/jwt.key` is missing, a key at `data/jwt.key`, then at the executable's directory, is copied there with mode 0600 and the old file is removed; otherwise a new key is generated. An existing key is never overwritten. Upgrading a Docker install signs every user out once, because the old key is in the container being replaced; to keep sessions, copy it out before recreating the container:
@@ -13,6 +18,11 @@ All notable changes to this project are documented in this file. The format is b
   docker cp artex-artex-1:/app/jwt.key ./keys/jwt.key
   ```
 
+### Security
+
+- **Login rate limit**: `POST /api/auth/login` allows 10 attempts per minute per client IP. The next attempt gets 429 with a `Retry-After` header (seconds) and a JSON error, and the IP is blocked for 1 minute. Each attempt while blocked, and each later block, doubles the wait, up to 15 minutes. A successful login clears the IP's count. The IP is the TCP peer address; `X-Forwarded-For` and similar headers are ignored, because the client controls them. There is no per-account lockout, so failing logins cannot lock the admin out from other addresses. The state is in memory: a restart clears it, an IP idle for 15 minutes is dropped, and at most 10,000 IPs are tracked (the least recently seen is dropped first). Behind a reverse proxy every client has the proxy's address and shares one budget, so an attacker failing logins through the proxy also blocks the real users behind it.
+- **`GET /api/health` hides the version from anonymous callers**: without a valid token it answers `{"ok":true}` only. With a token (the web UI sends one for the version in its top bar) the answer is unchanged: `ok`, `service` and `version`.
+
 ## [0.5.0] - 2026-11-04
 
 Ollama support, a setup token for the first-run claim, secret scrubbing for child processes, a Docker image that builds from a fresh clone, Go 1.26.10 and Node 22, and race and shellcheck runs in CI. Entries marked "direct commit" landed on `main` without a pull request.

cmd/artex/main.go

@@ -50,7 +50,7 @@ func main() {
 
 func run() int {
 	var (
-		addr    = flag.String("addr", ":8787", "HTTP listen address")
+		addr    = flag.String("addr", "127.0.0.1:8787", "HTTP listen address")
 		dataDir = flag.String("data", filepath.Join(config.BaseDir(), "data"), "data directory for SQLite stores (default: data/ next to the executable)")
 		proxy   = flag.String("proxy", "127.0.0.1:8788", "traffic recording proxy address (empty to disable)")
 	)

server/auth.go

@@ -10,6 +10,7 @@ import (
 	"net/http"
 	"os"
 	"path/filepath"
+	"strconv"
 	"strings"
 	"time"
 	"unicode/utf8"
@@ -29,7 +30,9 @@ const (
 	// The lower bound matches the setup page's frontend validation — validation that
 	// lives only in the frontend is as good as none: hitting the API directly bypasses
 	// it. The upper bound is bcrypt's hard limit: past 72 bytes GenerateFromPassword returns ErrPasswordTooLong; rejecting early beats a cryptic "password encryption failed".
-	minPasswordRunes = 8
+	// Both apply to new passwords only (init, change-password); login never checks
+	// length, so passwords set under the old 8-character minimum still log in.
+	minPasswordRunes = 12
 	maxPasswordBytes = 72
 )
 
@@ -372,6 +375,21 @@ func (s *Server) authLogin(w http.ResponseWriter, r *http.Request) {
 	if pg == nil {
 		return
 	}
+	s.login(w, r, pg.GetSetting)
+}
+
+// login is authLogin past the data-source check; getSetting is pg.GetSetting,
+// a parameter so tests can run it without PostgreSQL. Every attempt counts
+// against the caller's IP (s.loginLimit) and a successful one clears it. There
+// is no per-account lockout and no password length check here.
+func (s *Server) login(w http.ResponseWriter, r *http.Request, getSetting func(string) (string, bool, error)) {
+	ip := peerIP(r)
+	if wait := s.loginLimit.allow(ip); wait > 0 {
+		secs := int((wait + time.Second - 1) / time.Second)
+		w.Header().Set("Retry-After", strconv.Itoa(secs))
+		writeErr(w, 429, fmt.Sprintf("Too many login attempts, try again in %d seconds", secs))
+		return
+	}
 	var req struct {
 		Username string `json:"username"`
 		Password string `json:"password"`
@@ -384,7 +402,7 @@ func (s *Server) authLogin(w http.ResponseWriter, r *http.Request) {
 		writeErr(w, 401, "Incorrect username or password")
 		return
 	}
-	hash, ok, err := pg.GetSetting(authPassKey)
+	hash, ok, err := getSetting(authPassKey)
 	if err != nil {
 		writeErr(w, 503, errDataSourceUnavailable)
 		return
@@ -397,6 +415,7 @@ func (s *Server) authLogin(w http.ResponseWriter, r *http.Request) {
 		writeErr(w, 401, "Incorrect username or password")
 		return
 	}
+	s.loginLimit.reset(ip)
 	tok, err := signJWT(s.jwtKey)
 	if err != nil {
 		writeErr(w, 500, "token generation failed")

server/auth_test.go

@@ -302,10 +302,11 @@ func TestValidatePassword(t *testing.T) {
 		wantErr  bool
 	}{
 		{"empty", "", true},
-		{"seven chars", "1234567", true},
-		{"eight chars", "12345678", false},
-		{"eight multibyte chars counted as runes not bytes", strings.Repeat("€", 8), false},
-		{"three chars reach 9 bytes but only 3 runes", strings.Repeat("€", 3), true},
+		{"eight chars, the old minimum", "12345678", true},
+		{"eleven chars", "12345678901", true},
+		{"twelve chars", "123456789012", false},
+		{"twelve multibyte chars counted as runes not bytes", strings.Repeat("€", 12), false},
+		{"four chars reach 12 bytes but only 4 runes", strings.Repeat("€", 4), true},
 		{"72 bytes", strings.Repeat("a", 72), false},
 		{"73 bytes exceeds the bcrypt cap", strings.Repeat("a", 73), true},
 	} {

server/login_limit.go

@@ -0,0 +1,126 @@
+package server
+
+import (
+	"net"
+	"net/http"
+	"sync"
+	"time"
+)
+
+// Login attempts are limited per client IP, never per account: a per-account
+// lockout would let anyone lock the admin out by failing on purpose.
+const (
+	loginMaxAttempts = 10               // attempts allowed per IP per window
+	loginWindow      = time.Minute      // the counting window
+	loginBaseBlock   = time.Minute      // length of the first block
+	loginMaxBlock    = 15 * time.Minute // cap on the block; an entry idle this long is dropped
+	loginSweepEvery  = time.Minute      // how often stale entries are swept
+	loginMaxEntries  = 10000            // hard cap on tracked IPs
+)
+
+type loginEntry struct {
+	attempts     int
+	windowStart  time.Time
+	blockedUntil time.Time
+	block        time.Duration // current block length; doubles on every new block or blocked attempt
+	lastSeen     time.Time
+}
+
+// loginLimiter counts login attempts per IP. Past loginMaxAttempts in one
+// loginWindow the IP is blocked for loginBaseBlock; every attempt while blocked,
+// and every later block, doubles the block up to loginMaxBlock. State is in
+// memory only and swept of idle entries, so a restart clears it.
+type loginLimiter struct {
+	mu        sync.Mutex
+	now       func() time.Time
+	entries   map[string]*loginEntry
+	lastSweep time.Time
+}
+
+func newLoginLimiter(now func() time.Time) *loginLimiter {
+	return &loginLimiter{now: now, entries: map[string]*loginEntry{}}
+}
+
+// allow records one attempt from ip. It returns 0 when the attempt may go on,
+// otherwise how long ip must wait.
+func (l *loginLimiter) allow(ip string) time.Duration {
+	l.mu.Lock()
+	defer l.mu.Unlock()
+	now := l.now()
+	l.sweep(now)
+	e := l.entries[ip]
+	if e == nil {
+		if len(l.entries) >= loginMaxEntries {
+			l.evictOldest()
+		}
+		e = &loginEntry{windowStart: now}
+		l.entries[ip] = e
+	}
+	e.lastSeen = now
+	if now.Before(e.blockedUntil) {
+		e.block = min(2*e.block, loginMaxBlock)
+		e.blockedUntil = now.Add(e.block)
+		return e.block
+	}
+	if now.Sub(e.windowStart) >= loginWindow {
+		e.attempts, e.windowStart = 0, now
+	}
+	if e.attempts >= loginMaxAttempts {
+		if e.block == 0 {
+			e.block = loginBaseBlock
+		} else {
+			e.block = min(2*e.block, loginMaxBlock)
+		}
+		e.blockedUntil = now.Add(e.block)
+		e.attempts, e.windowStart = 0, now
+		return e.block
+	}
+	e.attempts++
+	return 0
+}
+
+// reset forgets ip, after a successful login. A blocked IP never reaches the
+// password check, so this only runs outside a block.
+func (l *loginLimiter) reset(ip string) {
+	l.mu.Lock()
+	defer l.mu.Unlock()
+	delete(l.entries, ip)
+}
+
+// sweep drops entries idle for loginMaxBlock, at most once per loginSweepEvery.
+// Such an entry is never blocked: a block ends at most loginMaxBlock after the
+// attempt that set it.
+func (l *loginLimiter) sweep(now time.Time) {
+	if now.Sub(l.lastSweep) < loginSweepEvery {
+		return
+	}
+	l.lastSweep = now
+	for ip, e := range l.entries {
+		if now.Sub(e.lastSeen) >= loginMaxBlock {
+			delete(l.entries, ip)
+		}
+	}
+}
+
+// evictOldest drops the least recently seen entry to stay under loginMaxEntries.
+func (l *loginLimiter) evictOldest() {
+	var oldest string
+	var at time.Time
+	for ip, e := range l.entries {
+		if oldest == "" || e.lastSeen.Before(at) {
+			oldest, at = ip, e.lastSeen
+		}
+	}
+	delete(l.entries, oldest)
+}
+
+// peerIP is the host part of the TCP peer address. X-Forwarded-For and similar
+// headers are ignored on purpose: the client sets them, so trusting them would
+// give an attacker a fresh bucket per request.
+func peerIP(r *http.Request) string {
+	host, _, err := net.SplitHostPort(r.RemoteAddr)
+	if err != nil {
+		return r.RemoteAddr
+	}
+	return host
+}

server/login_limit_test.go

@@ -0,0 +1,210 @@
+package server
+
+import (
+	"encoding/json"
+	"fmt"
+	"net/http/httptest"
+	"strings"
+	"testing"
+	"time"
+
+	"golang.org/x/crypto/bcrypt"
+)
+
+type fakeClock struct{ t time.Time }
+
+func (c *fakeClock) now() time.Time          { return c.t }
+func (c *fakeClock) advance(d time.Duration) { c.t = c.t.Add(d) }
+
+func newFakeClock() *fakeClock { return &fakeClock{t: time.Date(2026, 1, 1, 0, 0, 0, 0, time.UTC)} }
+
+// oldPassword is 8 characters: valid under the old minimum, too short for a new password now.
+const oldPassword = "abcd1234"
+
+// loginServer returns a Server whose login reads a stored hash of oldPassword
+// from a stub instead of PostgreSQL.
+func loginServer(t *testing.T, clk *fakeClock) (*Server, func(remote, password string, hdr map[string]string) *httptest.ResponseRecorder) {
+	t.Helper()
+	hash, err := bcrypt.GenerateFromPassword([]byte(oldPassword), bcrypt.MinCost)
+	if err != nil {
+		t.Fatal(err)
+	}
+	getSetting := func(key string) (string, bool, error) {
+		if key != authPassKey {
+			return "", false, nil
+		}
+		return string(hash), true, nil
+	}
+	s := &Server{jwtKey: []byte("0123456789abcdef0123456789abcdef"), loginLimit: newLoginLimiter(clk.now)}
+	post := func(remote, password string, hdr map[string]string) *httptest.ResponseRecorder {
+		body := fmt.Sprintf(`{"username":"ARTEX","password":%q}`, password)
+		r := httptest.NewRequest("POST", "/api/auth/login", strings.NewReader(body))
+		r.RemoteAddr = remote
+		for k, v := range hdr {
+			r.Header.Set(k, v)
+		}
+		w := httptest.NewRecorder()
+		s.login(w, r, getSetting)
+		return w
+	}
+	return s, post
+}
+
+func failN(t *testing.T, post func(string, string, map[string]string) *httptest.ResponseRecorder, remote string, n int) {
+	t.Helper()
+	for i := 0; i < n; i++ {
+		if w := post(remote, "wrong-password", nil); w.Code != 401 {
+			t.Fatalf("attempt %d from %s: status=%d want 401; body=%s", i+1, remote, w.Code, w.Body.String())
+		}
+	}
+}
+
+func TestLoginRateLimitReturns429WithRetryAfter(t *testing.T) {
+	clk := newFakeClock()
+	_, post := loginServer(t, clk)
+	const ip = "192.0.2.10:40000"
+	failN(t, post, ip, loginMaxAttempts)
+
+	w := post(ip, oldPassword, nil) // even the right password is refused while blocked
+	if w.Code != 429 {
+		t.Fatalf("status=%d want 429; body=%s", w.Code, w.Body.String())
+	}
+	if got := w.Header().Get("Retry-After"); got != "60" {
+		t.Fatalf("Retry-After=%q want 60", got)
+	}
+	var body struct {
+		Error string `json:"error"`
+		Token string `json:"token"`
+	}
+	if err := json.Unmarshal(w.Body.Bytes(), &body); err != nil || body.Error == "" || body.Token != "" {
+		t.Fatalf("want a JSON error body without a token, got %s (err=%v)", w.Body.String(), err)
+	}
+
+	// Each attempt while blocked doubles the wait, capped at loginMaxBlock.
+	for _, want := range []string{"120", "240", "480", "900", "900"} {
+		if got := post(ip, "wrong-password", nil).Header().Get("Retry-After"); got != want {
+			t.Fatalf("Retry-After=%q want %s", got, want)
+		}
+	}
+
+	// Once the block runs out the IP gets a fresh window.
+	clk.advance(loginMaxBlock)
+	if w := post(ip, oldPassword, nil); w.Code != 200 {
+		t.Fatalf("after the block: status=%d want 200; body=%s", w.Code, w.Body.String())
+	}
+}
+
+func TestLoginSuccessResetsCounter(t *testing.T) {
+	clk := newFakeClock()
+	_, post := loginServer(t, clk)
+	const ip = "192.0.2.11:40000"
+	failN(t, post, ip, loginMaxAttempts-1)
+	if w := post(ip, oldPassword, nil); w.Code != 200 {
+		t.Fatalf("status=%d want 200; body=%s", w.Code, w.Body.String())
+	}
+	// The counter starts over: a full budget of failures before the next block.
+	failN(t, post, ip, loginMaxAttempts)
+	if w := post(ip, "wrong-password", nil); w.Code != 429 {
+		t.Fatalf("status=%d want 429", w.Code)
+	}
+}
+
+func TestLoginRateLimitIsPerIP(t *testing.T) {
+	clk := newFakeClock()
+	_, post := loginServer(t, clk)
+	failN(t, post, "192.0.2.12:40000", loginMaxAttempts)
+	if w := post("192.0.2.12:40001", "wrong-password", nil); w.Code != 429 {
+		t.Fatalf("same IP, other port: status=%d want 429", w.Code)
+	}
+	if w := post("198.51.100.7:40000", "wrong-password", nil); w.Code != 401 {
+		t.Fatalf("other IP: status=%d want 401", w.Code)
+	}
+	if w := post("[2001:db8::1]:40000", oldPassword, nil); w.Code != 200 {
+		t.Fatalf("other IP: status=%d want 200", w.Code)
+	}
+}
+
+func TestLoginRateLimitIgnoresForwardedFor(t *testing.T) {
+	clk := newFakeClock()
+	_, post := loginServer(t, clk)
+	const blocked = "192.0.2.13"
+	failN(t, post, blocked+":40000", loginMaxAttempts)
+
+	// A blocked peer cannot escape by claiming another address.
+	for _, h := range []string{"X-Forwarded-For", "X-Real-IP"} {
+		if w := post(blocked+":40000", oldPassword, map[string]string{h: "203.0.113.9"}); w.Code != 429 {
+			t.Fatalf("%s from blocked peer: status=%d want 429", h, w.Code)
+		}
+	}
+	// And another peer naming the blocked address is not blocked.
+	if w := post("198.51.100.8:40000", oldPassword, map[string]string{"X-Forwarded-For": blocked}); w.Code != 200 {
+		t.Fatalf("X-Forwarded-For naming the blocked IP: status=%d want 200", w.Code)
+	}
+}
+
+// The 12-character minimum applies to new passwords only. A hash stored under the
+// old 8-character minimum must still log in.
+func TestEightCharPasswordLogsInButIsRejectedAsNew(t *testing.T) {
+	if n := len([]rune(oldPassword)); n != 8 {
+		t.Fatalf("oldPassword has %d characters, want 8", n)
+	}
+	_, post := loginServer(t, newFakeClock())
+	w := post("192.0.2.14:40000", oldPassword, nil)
+	if w.Code != 200 || !strings.Contains(w.Body.String(), `"token"`) {
+		t.Fatalf("login with an existing 8-character password: status=%d body=%s", w.Code, w.Body.String())
+	}
+	// authInit and authChangePassword both refuse a new password through validatePassword.
+	if msg := validatePassword(oldPassword); msg == "" {
+		t.Fatal("validatePassword accepted an 8-character new password")
+	}
+}
+
+func TestLoginLimiterSweepsIdleEntries(t *testing.T) {
+	clk := newFakeClock()
+	l := newLoginLimiter(clk.now)
+	l.allow("192.0.2.20")
+	clk.advance(loginMaxBlock - time.Second)
+	l.allow("192.0.2.21") // sweeps, but .20 is not idle long enough yet
+	if len(l.entries) != 2 {
+		t.Fatalf("entries=%d want 2 before either is idle for %v", len(l.entries), loginMaxBlock)
+	}
+	clk.advance(loginSweepEvery)
+	l.allow("192.0.2.22") // next sweep drops .20
+	if _, ok := l.entries["192.0.2.20"]; ok || len(l.entries) != 2 {
+		t.Fatalf("idle entry not swept: %d entries", len(l.entries))
+	}
+}
+
+func TestLoginLimiterCapsEntries(t *testing.T) {
+	l := newLoginLimiter(newFakeClock().now)
+	for i := 0; i <= loginMaxEntries; i++ {
+		l.allow(fmt.Sprintf("10.%d.%d.%d", i>>16&255, i>>8&255, i&255))
+	}
+	if len(l.entries) != loginMaxEntries {
+		t.Fatalf("entries=%d want the cap %d", len(l.entries), loginMaxEntries)
+	}
+}
+
+func TestHealthHidesVersionWithoutToken(t *testing.T) {
+	key := []byte("0123456789abcdef0123456789abcdef")
+	s := &Server{jwtKey: key}
+
+	w := httptest.NewRecorder()
+	s.health(w, httptest.NewRequest("GET", "/api/health", nil))
+	if got := strings.TrimSpace(w.Body.String()); w.Code != 200 || got != `{"ok":true}` {
+		t.Fatalf("no token: status=%d body=%s want {\"ok\":true}", w.Code, got)
+	}
+
+	tok, err := signJWT(key)
+	if err != nil {
+		t.Fatal(err)
+	}
+	r := httptest.NewRequest("GET", "/api/health", nil)
+	r.Header.Set("Authorization", "Bearer "+tok)
+	w = httptest.NewRecorder()
+	s.health(w, r)
+	var body map[string]any
+	if err := json.Unmarshal(w.Body.Bytes(), &body); err != nil || body["version"] != BuildVersion || body["service"] != "artex" {
+		t.Fatalf("valid token: body=%s want service and version", w.Body.String())
+	}
+}

server/server.go

@@ -51,6 +51,8 @@ type Server struct {
 	setupMu    sync.Mutex
 	setupToken string // one-time token POST /api/auth/init requires; empty when none is active
 
+	loginLimit *loginLimiter // per-IP attempt limit on POST /api/auth/login
+
 	// concMu serializes concurrency-cap decisions (admission + reconcile) so a
 	// scheduler tick and an HTTP settings change / task creation can't both count
 	// free slots off the same snapshot and over-promote past the limit.
@@ -155,7 +157,8 @@ func New(ctx context.Context, m *Manager, skillDir string, dataDir string, keyDi
 		triggerActive: map[string]int{}, triggerCfg: map[string]triggerBehavior{},
 		profChatAgents: map[int64]*agent.ChatAgent{},
 		provByProfile:  map[int64]*provEntry{}, llmHealth: newLLMHealthRegistry(m.pg),
-		taskAgents: map[string]*taskAgentBundle{}, archiveWake: make(chan struct{}, 1)}
+		taskAgents: map[string]*taskAgentBundle{}, archiveWake: make(chan struct{}, 1),
+		loginLimit: newLoginLimiter(time.Now)}
 	for _, o := range opts {
 		o(s)
 	}
@@ -964,7 +967,14 @@ func (s *Server) Handler() http.Handler {
 
 // --- handlers ---
 
+// health answers without a token (requireAuth exempts it), but only a caller
+// with a valid token gets the service name and version; the web UI's top bar
+// reads the version, and sends its token.
 func (s *Server) health(w http.ResponseWriter, r *http.Request) {
+	if !verifyJWT(extractToken(r), s.jwtKey) {
+		writeJSON(w, 200, map[string]any{"ok": true})
+		return
+	}
 	writeJSON(w, 200, map[string]any{"ok": true, "service": "artex", "version": BuildVersion})
 }
 

web/src/app/(auth)/setup/page.tsx

@@ -41,8 +41,8 @@ export default function SetupPage() {
       setError("The two passwords do not match");
       return;
     }
-    if (password.length < 8) {
-      setError("Password must be at least 8 characters");
+    if (password.length < 12) {
+      setError("Password must be at least 12 characters");
       return;
     }
     setLoading(true);
@@ -81,7 +81,7 @@ export default function SetupPage() {
             <p className="mx-auto max-w-xl text-muted-foreground">
               {unavailable
                 ? "The backend or database is temporarily unavailable. To avoid overwriting a password already set on this instance, setup is temporarily closed — restore the service and try again."
-                : "First time using ARTEX? Enter the setup token from the server log, then set a login password for the account (at least 8 characters)"}
+                : "First time using ARTEX? Enter the setup token from the server log, then set a login password for the account (at least 12 characters)"}
             </p>
           </div>
           {unavailable ? (
@@ -111,7 +111,7 @@ export default function SetupPage() {
                   type="password"
                   value={password}
                   onChange={(e) => setPassword(e.target.value)}
-                  placeholder="At least 8 characters"
+                  placeholder="At least 12 characters"
                   autoComplete="new-password"
                 />
               </div>