fix(auth): make the published bind address configurable

Publishing 127.0.0.1:8787 removed the only working access path for installs
reached through a host or WSL IP, without an obvious way back. Keep loopback as
the default and let ARTEX_BIND_ADDR override it (0.0.0.0 restores the old
behaviour; the setup token still protects the first-run claim).

.env.example

@@ -1,5 +1,8 @@
 # Docker deployment configuration (copy to .env)
-ARTEX_TAG=latest
+# The image is built locally by `docker compose build`; there is no registry tag.
+# Address the web UI is published on. 127.0.0.1 keeps it host-local; set 0.0.0.0
+# to reach it from another machine or from Windows when ARTEX runs inside WSL.
+ARTEX_BIND_ADDR=127.0.0.1
 POSTGRES_USER=artex
 POSTGRES_PASSWORD=change-me-please
 POSTGRES_DB=artex

CHANGELOG.md

@@ -4,13 +4,26 @@ All notable changes to this project are documented in this file. The format is b
 
 ## [Unreleased]
 
+### Changed
+
+- **The image is built from the checkout instead of pulled**: `docker-compose.yml` no longer references the `autumn27/artex` namespace, whose repository is gone and whose name its holder could republish at any time. `install.sh` and `update.sh` build the image locally.
+- **`update.sh` refuses to rebuild stale code**: when `git pull --ff-only` fails on a diverged clone (including every clone made before the 2026-10-23 history rewrite), it now stops with recovery commands instead of warning, rebuilding the old tree and printing "Build complete".
+
+### Fixed
+
+- **Ollama base URL on the environment path**: `ARTEX_LLM_BASE_URL=http://host:11434` is normalized to the `/v1` root the same way the saved-profile path already was, so a LAN Ollama server configured through the environment no longer answers 404 on chat.
+
+### Added
+
+- **Ollama support**: an `ollama` LLM profile format for a local or LAN Ollama server. Keyless by design: no API key is needed, `OPENAI_API_KEY` is never inherited, the base URL defaults to `http://127.0.0.1:11434/v1` and is normalized to the `/v1` root, and the profile works in the failover pool, task LLM chains, model listing and connection tests. Contributed by @glauberportella (#128).
+
 ## [0.4.0] - 2026-10-23
 
 First release of this fork. The engine underneath is still upstream's, but this tree carries a long stretch of operational work on top of it: a recon toolbox and skills, findings discipline, cost accounting, and container packaging.
 
 ### Added
 
-- **Recon toolbox in the image**: pinned nuclei, httpx, subfinder, katana, ffuf, dalfox and a sqlmap checkout, verified at build time, so hunt workers can call them directly (#112).
+- **Recon toolbox in the image**: nuclei, httpx, subfinder, katana, ffuf and dalfox installed at release-pinned versions, plus a sqlmap checkout that tracks the default branch, each smoke-tested at build time (run once, not checksum-verified), so hunt workers can call them directly (#112). Checksum verification arrives with #127.
 - **Recursive recon skill** (`bbot-recon`): bbot v3.0.2 in its own virtualenv with module dependencies preinstalled. Policy-gated modes (passive seeds-only when scanners are banned, active-safe, loud only on explicit permission) and JSONL ingestion into the facts graph (#118).
 - **php2shell skill**: turns a file-inclusion finding into verified RCE through the session.upload_progress chain, lab first and gated on program policy (#113).
 - **Code audit and reverse-engineering skills** (#55, #77).
@@ -30,9 +43,10 @@ First release of this fork. The engine underneath is still upstream's, but this
 - **Planner quiet mode** for less narration during long runs (#81, #88).
 - **Findings discipline**: proof-basis gates, measured impact, adversarial verification and dedup rules decide what a finding must contain before it is reported (#51, #53, #63, #69).
 - **Prompt-injection hardening** on seed material and MCP tool descriptions (#49).
-- **Pause preserves the task timeout budget**, and paused time is frozen in the task clock (#2, #65, #110).
+- **Pause preserves the task timeout budget**, and paused time is frozen in the task clock (#2, #110).
+- **Deadline contract** in the executor seeds (#65).
 - **Task detail metrics and outcome cards** summarise a finished run (#92, #102).
-- **Skill-gap, loop-contract and shadow-lesson reporting** for the planner (#104, #106, #108).
+- **Skill-gap and loop-contract reporting** for the planner (#104, #106); the RSI notes land as documentation in #108.
 
 ### Removed
 

README.md

@@ -84,10 +84,13 @@ cd ARTEX
 
 The script will: detect / auto-install Docker → then let you choose **① all-Docker** or **② local build and run**:
 
-- **① All-Docker**: enter a Postgres password (press Enter for a random one) → auto-writes `.env` → `docker compose up -d`.
+- **① All-Docker**: enter a Postgres password (press Enter for a random one) → auto-writes `.env` → `docker compose up -d --build` (the image is built from this checkout).
 - **② Local run**: choose a database (connect to an existing one / start one with Docker) → generates `config.json` → `go` builds the embedded single binary → starts.
 
 Once installed, open **http://localhost:8787** (on first entry, `/setup` sets the admin password).
+`/setup` asks for a one-time setup token: while no password is set, the backend prints it at startup in its log (`docker compose logs artex | grep "Setup token"`, or the terminal running `./start.sh`).
+
+The web UI is published on `127.0.0.1` by default. To reach it from another machine, or from Windows when ARTEX runs inside WSL, set `ARTEX_BIND_ADDR=0.0.0.0` in `.env` (the setup token still protects the first-run claim), or keep loopback and use a reverse proxy or an SSH tunnel.
 
 ### Option 2: Docker Compose (manual)
 
@@ -95,10 +98,14 @@ Once installed, open **http://localhost:8787** (on first entry, `/setup` sets th
 git clone https://github.com/BaaRai/ARTEX.git
 cd ARTEX
 cp .env.example .env          # set POSTGRES_PASSWORD, optionally ANTHROPIC_API_KEY
-docker compose up -d          # pulls the autumn27/artex image + postgres
+# build the single binary first (frontend, then go build -tags embedui), see the Dockerfile header
+docker compose up -d --build  # builds artex from this checkout + starts postgres
 # → http://localhost:8787
 ```
 
+> The image is built locally. There is no published image to pull: the previous
+> default pointed at a third-party namespace that no longer hosts this project.
+
 The image already includes common tools (ripgrep/curl/vim/npm/nmap…); `./skills` and `./data` are persisted via bind mounts.
 
 For remote MCP, you can choose `http` (Streamable HTTP) or `sse` (legacy SSE) in system settings.
@@ -175,6 +182,19 @@ Daily cron example (03:20, operator crontab):
 
 > Upgrading only swaps the program and never touches data: the Postgres volume `pgdata`, `./data` (jwt.key / SQLite, etc.), and `./skills` are all preserved. **No manual database migration is needed**: `artex` idempotently re-runs `schema.sql` on every start (with `ADD COLUMN` / `CREATE INDEX IF NOT EXISTS`), i.e. "restart equals migrate". It is still recommended to back up `./data` and the database before upgrading.
 
+### Upgrading from a clone made before 2026-10-23
+
+The repository history was rewritten that day to drop a personal email address from the commit metadata. A clone made before then cannot fast-forward, and `./update.sh` now stops instead of silently rebuilding stale code. Recover once, then upgrade normally:
+
+```bash
+git stash                 # only if you have local changes
+git fetch origin
+git reset --hard origin/main
+git stash pop             # only if you stashed
+```
+
+Your data is not affected: `./data`, the `pgdata` volume and `./skills` are not part of the git history.
+
 ### Option 1: one-click update script
 
 ```bash
@@ -184,17 +204,16 @@ cd ARTEX
 
 The script first optionally `git pull`s the latest code, then lets you choose **① Docker update** or **② local build update** (matching `install.sh`):
 
-- **① Docker**: you can specify a target image tag (press Enter to keep `.env`'s `ARTEX_TAG`, default `latest`) → `docker compose pull` → `docker compose up -d` (restarting with the new image migrates automatically).
+- **① Docker**: rebuilds the image from this checkout → `docker compose build artex` → `docker compose up -d artex` (restarting migrates the schema automatically).
 - **② Local**: rebuild the frontend static assets → recompile `./artex` (restart the process for it to take effect).
 
 ### Option 2: Docker Compose (manual)
 
 ```bash
 cd ARTEX
-git pull                       # update compose / scripts (optional)
-# pin a version: set ARTEX_TAG=v0.2.0 in .env; otherwise latest is used
-docker compose pull artex
-docker compose up -d artex     # restart with the new image → schema migrates automatically
+git pull                       # update compose / scripts
+docker compose build artex     # rebuild the image from this checkout
+docker compose up -d artex     # recreate the container → schema migrates automatically
 docker image prune -f          # clean up old images (optional)
 ```
 
@@ -233,8 +252,8 @@ CGO_ENABLED=0 go build -tags embedui -o artex ./cmd/artex
 }
 ```
 
-**LLM**: `export ANTHROPIC_API_KEY=sk-...` (or `OPENAI_API_KEY`); you can also fill it in on the UI's "LLM Configuration" page.
-Optional: `ARTEX_LLM_PROVIDER` / `ARTEX_LLM_MODEL` / `ARTEX_LLM_BASE_URL` / `ARTEX_LLM_PROXY`.
+**LLM**: `export ANTHROPIC_API_KEY=sk-...` (or `OPENAI_API_KEY`); you can also fill it in on the UI's "LLM Configuration" page — which can add a local Ollama server as a keyless provider (format "Ollama", base URL default `http://127.0.0.1:11434/v1`, models loaded from the server).
+Optional: `ARTEX_LLM_PROVIDER` (anthropic | openai | openai-responses | ollama) / `ARTEX_LLM_MODEL` / `ARTEX_LLM_BASE_URL` / `ARTEX_LLM_PROXY`.
 
 **Concurrency**: the number of work agents per task is configured in "System Settings" (default 3).
 

agent/provider.go

@@ -3,8 +3,10 @@
 // §4.3 (planner) and §4.4 (work agent).
 //
 // Provider configuration is read from the environment so the system runs with
-// any Anthropic- or OpenAI-format endpoint. If no key is configured, FromEnv
-// returns ok=false and the exploration engine stays idle (an LLM is required).
+// any Anthropic-, OpenAI- or Ollama-format endpoint (the last being keyless).
+// If no usable configuration is found — a key for the cloud formats, a base
+// URL for keyless Ollama — FromEnv returns ok=false and the exploration engine
+// stays idle (an LLM is required).
 package agent
 
 import (
@@ -125,6 +127,47 @@ const (
 	maxWindowK     = 1000 // cap at 1M tokens (user request)
 )
 
+// FormatOllama marks a LOCAL Ollama server profile. Ollama exposes an
+// OpenAI-compatible Chat Completions endpoint (/v1), so on the wire it speaks
+// llm.FormatOpenAI; the distinct sentinel matters everywhere a profile must be
+// treated as usable WITHOUT an API key — Ollama runs keyless on localhost/LAN.
+// See NeedsAPIKey and NewProvider (which maps it onto FormatOpenAI while
+// forcing our base-URL/key defaults so an empty key never falls back to
+// OpenAI's public endpoint).
+const FormatOllama llm.Format = "ollama"
+
+// DefaultOllamaBaseURL is the base of Ollama's OpenAI-compatible endpoint the
+// provider appends /chat/completions to. Must never end up empty: with an
+// empty BaseURL, the SDK would fall back to OpenAI's public endpoint.
+const DefaultOllamaBaseURL = "http://127.0.0.1:11434/v1"
+
+// ollamaPlaceholderKey is sent as the Bearer token when no key is configured.
+// Ollama ignores authorization, but a concrete value keeps the SDK from
+// falling back to OPENAI_API_KEY (which would silently ship a real cloud key
+// to the local server) and keeps the header stable.
+const ollamaPlaceholderKey = "ollama"
+
+// normalizeOllamaBaseURL maps whatever the operator typed onto the /v1 root the
+// OpenAI-compatible provider expects: it strips a pasted "/chat/completions"
+// suffix, an empty value becomes the local default, and a missing "/v1" is
+// appended so the natural "http://host:11434" also works. Shared by the saved
+// profile path (ConfigFrom) and the environment path (FromEnv): both used to
+// differ, and only the profile path normalized.
+func normalizeOllamaBaseURL(raw string) string {
+	b := strings.TrimRight(strings.TrimSuffix(strings.TrimSpace(raw), "/chat/completions"), "/")
+	if b == "" {
+		return DefaultOllamaBaseURL
+	}
+	if !strings.HasSuffix(b, "/v1") {
+		b += "/v1"
+	}
+	return b
+}
+
+// NeedsAPIKey reports whether this provider format requires an API key to be
+// usable. Ollama (local server) is keyless; every cloud format needs one.
+func (c Config) NeedsAPIKey() bool { return c.Format != FormatOllama }
+
 // CompactionWindow returns the model context window in TOKENS for compaction
 // thresholds, resolved from the user-configured size (ContextWindowK). 0/unset →
 // a 200K default; otherwise clamped to [32K, 1M] so compaction stays effective.
@@ -154,11 +197,11 @@ func compactionConfig(windowTokens int) *compaction.Config {
 
 // FromEnv resolves the LLM provider config:
 //
-//	ARTEX_LLM_PROVIDER = anthropic|openai (default: inferred from keys)
+//	ARTEX_LLM_PROVIDER = anthropic|openai|openai-responses|ollama (default: inferred from keys)
 //	ARTEX_LLM_MODEL    = model id        (default: per provider)
-//	ARTEX_LLM_BASE_URL = endpoint        (optional)
+//	ARTEX_LLM_BASE_URL = endpoint        (optional; ollama defaults to localhost:11434/v1)
 //	ARTEX_LLM_PROXY    = proxy URL        (optional; http/https/socks5)
-//	ANTHROPIC_API_KEY / OPENAI_API_KEY         = credentials
+//	ANTHROPIC_API_KEY / OPENAI_API_KEY         = credentials (unused for ollama)
 func FromEnv() (Config, bool) {
 	prov := os.Getenv("ARTEX_LLM_PROVIDER")
 	anthKey := os.Getenv("ANTHROPIC_API_KEY")
@@ -196,6 +239,16 @@ func FromEnv() (Config, bool) {
 		if c.Model == "" {
 			c.Model = "gpt-5"
 		}
+	case "ollama":
+		c.Format = FormatOllama
+		// Deliberately NOT inheriting OPENAI_API_KEY: a local Ollama ignores the
+		// bearer token, so the only thing the inheritance could do is ship a real
+		// cloud key to the endpoint. A gateway that does require a key can still
+		// be configured as a saved profile with an explicit one.
+		c.BaseURL = normalizeOllamaBaseURL(c.BaseURL)
+		if c.Model == "" {
+			c.Model = "llama3.2"
+		}
 	default:
 		c.Format = llm.FormatAnthropic
 		c.APIKey = anthKey
@@ -203,7 +256,8 @@ func FromEnv() (Config, bool) {
 			c.Model = "claude-opus-4-8"
 		}
 	}
-	if c.APIKey == "" {
+	// Ollama runs without a key: only the cloud formats gate on one.
+	if c.APIKey == "" && c.NeedsAPIKey() {
 		return Config{}, false
 	}
 	return c, true
@@ -236,6 +290,16 @@ func ConfigFrom(provider, model, baseURL, apiKey, proxy string) Config {
 		if c.Model == "" {
 			c.Model = "gpt-5"
 		}
+	case "ollama":
+		c.Format = FormatOllama
+		// Ollama's OpenAI-compatible endpoint lives at /v1 and the provider
+		// appends "/chat/completions". Never leave it empty: an empty BaseURL
+		// would let the SDK fall back to OpenAI's public endpoint with this
+		// profile's request.
+		c.BaseURL = normalizeOllamaBaseURL(c.BaseURL)
+		if c.Model == "" {
+			c.Model = "llama3.2"
+		}
 	default:
 		c.Format = llm.FormatAnthropic
 		// provider appends "/v1/messages".
@@ -257,13 +321,16 @@ func isFalsy(s string) bool {
 	return false
 }
 
-// Provider returns the short provider name ("anthropic"/"openai").
+// Provider returns the short provider name
+// ("anthropic"/"openai"/"openai-responses"/"ollama").
 func (c Config) Provider() string {
 	switch c.Format {
 	case llm.FormatOpenAI:
 		return "openai"
 	case llm.FormatOpenAIResponses:
 		return "openai-responses"
+	case FormatOllama:
+		return "ollama"
 	}
 	return "anthropic"
 }
@@ -276,10 +343,27 @@ func (c Config) NewProvider() (llm.Provider, error) {
 	if err != nil {
 		return nil, err
 	}
+	wireFormat := c.Format
+	baseURL := c.BaseURL
+	apiKey := c.APIKey
+	// Ollama speaks OpenAI's Chat Completions on the wire. Bundle the keyless
+	// defaults here so an Ollama config can never leak toward the SDK's env-
+	// driven fallbacks: BaseURL to OpenAI's public endpoint, APIKey to
+	// OPENAI_API_KEY. A local Ollama ignores the placeholder token; a gateway
+	// that requires one carries a real key in the profile.
+	if wireFormat == FormatOllama {
+		wireFormat = llm.FormatOpenAI
+		if strings.TrimSpace(baseURL) == "" {
+			baseURL = DefaultOllamaBaseURL
+		}
+		if strings.TrimSpace(apiKey) == "" {
+			apiKey = ollamaPlaceholderKey
+		}
+	}
 	lc := llm.Config{
-		Format:     c.Format,
-		BaseURL:    c.BaseURL,
-		APIKey:     c.APIKey,
+		Format:     wireFormat,
+		BaseURL:    baseURL,
+		APIKey:     apiKey,
 		Model:      c.Model,
 		HTTPClient: client,
 	}

cmd/artex/main.go

@@ -93,7 +93,7 @@ func run() int {
 		skillDir = abs
 	}
 	log.Printf("[config] skill directory: %s", skillDir)
-	srv := server.New(ctx, mgr, skillDir, *dataDir, config.BaseDir())
+	srv := server.New(ctx, mgr, skillDir, *dataDir, config.BaseDir(), server.WithSecretScrub())
 	httpSrv := &http.Server{
 		Addr:              *addr,
 		Handler:           srv.Handler(),

cmd/artex/main_test.go

@@ -7,6 +7,8 @@ import (
 	"github.com/Autumn-27/artex/agent"
 )
 
+// The shutdown cause must survive the parent context being cancelled first;
+// losing it would drop the diagnostic on every graceful stop.
 func TestShutdownContextPreservesNamedCause(t *testing.T) {
 	signalCtx, signalCancel := context.WithCancel(context.Background())
 	ctx, shutdown := shutdownContext(signalCtx)

db/config.go

@@ -143,13 +143,14 @@ func (d *DB) ProfileByID(id int64) (*LLMProfile, error) {
 }
 
 // PoolProfiles returns the failover chain in run order, api keys included: the
-// active profile first, then every other keyed profile that isn't excluded, by
+// active profile first, then every other usable profile that isn't excluded, by
 // priority DESC (id ASC to stay stable). Profiles without an api key can't serve
-// a request, so they never enter the chain. The ordering IS the policy — callers
-// walk the slice front to back.
+// a request — except Ollama ('ollama'), which is a keyless local server — so
+// they never enter the chain. The ordering IS the policy — callers walk the
+// slice front to back.
 func (d *DB) PoolProfiles() ([]*LLMProfile, error) {
 	rows, err := d.Query(`SELECT ` + profileColsKey + ` FROM llm_profiles
-WHERE COALESCE(api_key,'') <> '' AND (is_default OR NOT pool_exclude)
+WHERE (COALESCE(api_key,'') <> '' OR format='ollama') AND (is_default OR NOT pool_exclude)
 ORDER BY is_default DESC, priority DESC, id ASC`)
 	if err != nil {
 		return nil, err

db/schema.sql

@@ -363,7 +363,7 @@ CREATE TABLE IF NOT EXISTS settings (
 CREATE TABLE IF NOT EXISTS llm_profiles (
     id               BIGSERIAL PRIMARY KEY,
     name             TEXT NOT NULL UNIQUE,
-    format           TEXT NOT NULL CHECK (format IN ('openai','anthropic','openai-responses')),
+    format           TEXT NOT NULL CHECK (format IN ('openai','anthropic','openai-responses','ollama')),
     base_url         TEXT,
     proxy            TEXT,
     model            TEXT NOT NULL,
@@ -416,11 +416,12 @@ ALTER TABLE llm_profiles ADD COLUMN IF NOT EXISTS priority     INTEGER NOT NULL
 ALTER TABLE llm_profiles ADD COLUMN IF NOT EXISTS pool_exclude BOOLEAN NOT NULL DEFAULT false;
 -- Streaming toggle; backfills old databases. Default true = keep existing streaming behavior; old configs upgrade transparently.
 ALTER TABLE llm_profiles ADD COLUMN IF NOT EXISTS streaming    BOOLEAN NOT NULL DEFAULT true;
--- Relax the format constraint to allow openai-responses (OpenAI Responses API); backfills old databases.
--- Runs at every startup, idempotent: drop the old CHECK first, then create the new three-value CHECK.
+-- Relax the format constraint to allow openai-responses (OpenAI Responses API) and
+-- ollama (local Ollama server, keyless OpenAI-compatible endpoint); backfills old databases.
+-- Runs at every startup, idempotent: drop the old CHECK first, then create the new four-value CHECK.
 ALTER TABLE llm_profiles DROP CONSTRAINT IF EXISTS llm_profiles_format_check;
 ALTER TABLE llm_profiles ADD  CONSTRAINT llm_profiles_format_check
-    CHECK (format IN ('openai','anthropic','openai-responses'));
+    CHECK (format IN ('openai','anthropic','openai-responses','ollama'));
 
 -- Output cap and its field name; backfills old databases. Defaults 0 / '' = send no cap and keep the max_tokens field name;
 -- old configs behave exactly as before.

docker-compose.yml

@@ -15,7 +15,14 @@ services:
     restart: unless-stopped
 
   artex:
-    image: autumn27/artex:${ARTEX_TAG:-latest}
+    # Built from this checkout. The previous reference was a third-party
+    # namespace whose repository is gone: whoever controls that name could
+    # publish the tag this file would pull.
+    build:
+      context: .
+      args:
+        ARTEX_VERSION: ${ARTEX_VERSION:-dev}
+    image: artex:local
     depends_on:
       postgres:
         condition: service_healthy
@@ -28,7 +35,7 @@ services:
       ARTEX_LLM_BASE_URL: ${ARTEX_LLM_BASE_URL:-}
       ARTEX_LLM_PROXY: ${ARTEX_LLM_PROXY:-}
     ports:
-      - "8787:8787"
+      - "${ARTEX_BIND_ADDR:-127.0.0.1}:8787:8787"
       - "127.0.0.1:8788:8788"
     volumes:
       # Bind the data directory (jwt.key, SQLite, etc.) to data/ in the project root,

install.sh

@@ -45,9 +45,11 @@ install_docker(){
   else
     info "reusing existing .env"
   fi
-  info "pulling images and starting..."
-  docker compose pull || true
-  docker compose up -d
+  info "building the image from this checkout and starting..."
+  if [ ! -f "dist/amd64/artex" ] && [ ! -f "dist/arm64/artex" ]; then
+    warn "no prebuilt binary under dist/ - build it first (frontend, then go build -tags embedui); see the Dockerfile header"
+  fi
+  docker compose up -d --build
   ok "started → http://localhost:8787"
   info "view logs: docker compose logs -f artex"
 }

server/auth.go

@@ -2,6 +2,8 @@ package server
 
 import (
 	"crypto/rand"
+	"crypto/subtle"
+	"encoding/hex"
 	"fmt"
 	"log"
 	"math/big"
@@ -154,7 +156,34 @@ func (s *Server) authStatus(w http.ResponseWriter, r *http.Request) {
 	writeJSON(w, 200, map[string]any{"initialized": hash != ""})
 }
 
+// initSetupToken generates the one-time setup token when no admin password is set
+// and writes it to the server log. Without it, anyone who can reach the port
+// before the operator could claim the instance through POST /api/auth/init.
+func (s *Server) initSetupToken() {
+	if s.m == nil || s.m.pg == nil {
+		return
+	}
+	if hash, _, err := s.m.pg.GetSetting(authPassKey); err == nil && hash != "" {
+		return
+	}
+	buf := make([]byte, 16)
+	if _, err := rand.Read(buf); err != nil {
+		log.Fatalf("[auth] generate setup token: %v", err)
+	}
+	s.setupToken = hex.EncodeToString(buf)
+	log.Printf("[auth] no admin password is set. Setup token for the /setup page: %s", s.setupToken)
+}
+
+// checkSetupToken compares got with the active setup token in constant time.
+// It fails when no token is active.
+func (s *Server) checkSetupToken(got string) bool {
+	s.setupMu.Lock()
+	defer s.setupMu.Unlock()
+	return s.setupToken != "" && subtle.ConstantTimeCompare([]byte(got), []byte(s.setupToken)) == 1
+}
+
 // POST /api/auth/init — sets the password for the first time; rejected if already set.
+// Requires the setup token from the server log.
 func (s *Server) authInit(w http.ResponseWriter, r *http.Request) {
 	pg := s.pg(w)
 	if pg == nil {
@@ -170,12 +199,18 @@ func (s *Server) authInit(w http.ResponseWriter, r *http.Request) {
 		return
 	}
 	var req struct {
-		Password string `json:"password"`
+		Password   string `json:"password"`
+		SetupToken string `json:"setup_token"`
 	}
 	if err := decode(r, &req); err != nil || req.Password == "" {
 		writeErr(w, 400, "Password cannot be empty")
 		return
 	}
+	// 403, not 401: the frontend treats any 401 as an expired session and redirects to /login.
+	if !s.checkSetupToken(req.SetupToken) {
+		writeErr(w, 403, "Invalid setup token. The server log prints it at startup while no password is set")
+		return
+	}
 	if msg := validatePassword(req.Password); msg != "" {
 		writeErr(w, 400, msg)
 		return
@@ -197,6 +232,9 @@ func (s *Server) authInit(w http.ResponseWriter, r *http.Request) {
 		writeErr(w, 403, "Password is already set")
 		return
 	}
+	s.setupMu.Lock()
+	s.setupToken = ""
+	s.setupMu.Unlock()
 	tok, err := signJWT(s.jwtKey)
 	if err != nil {
 		writeErr(w, 500, "token generation failed")

server/auth_test.go

@@ -62,6 +62,62 @@ func TestAuthInitFailsClosedWhenDataSourceUnavailable(t *testing.T) {
 	}
 }
 
+// While no password is set, authInit must reject a missing or wrong setup token
+// and accept the one generated at startup, once.
+func TestAuthInitRequiresSetupToken(t *testing.T) {
+	m, err := NewManager(t.TempDir(), "")
+	if err != nil {
+		t.Skipf("postgres unavailable (%v)", err)
+	}
+	defer m.Close()
+	clearPassword := func() {
+		if _, err := m.pg.Exec(`DELETE FROM settings WHERE key=$1`, authPassKey); err != nil {
+			t.Fatal(err)
+		}
+	}
+	clearPassword()
+	defer clearPassword()
+
+	s := &Server{m: m, jwtKey: []byte("0123456789abcdef0123456789abcdef")}
+	s.initSetupToken()
+	setupToken := s.setupToken
+	if setupToken == "" {
+		t.Fatal("no setup token generated while no password is set")
+	}
+	post := func(body string) *httptest.ResponseRecorder {
+		w := httptest.NewRecorder()
+		s.authInit(w, httptest.NewRequest("POST", "/api/auth/init", strings.NewReader(body)))
+		return w
+	}
+
+	for name, body := range map[string]string{
+		"missing": `{"password":"correct horse battery"}`,
+		"empty":   `{"password":"correct horse battery","setup_token":""}`,
+		"wrong":   `{"password":"correct horse battery","setup_token":"0000"}`,
+	} {
+		if w := post(body); w.Code != 403 || strings.Contains(w.Body.String(), `"token"`) {
+			t.Fatalf("%s setup token: status=%d want 403 without a token; body=%s", name, w.Code, w.Body.String())
+		}
+	}
+	if hash, _, err := m.pg.GetSetting(authPassKey); err != nil || hash != "" {
+		t.Fatalf("a rejected request changed the password state: hash set=%v err=%v", hash != "", err)
+	}
+
+	ok := `{"password":"correct horse battery","setup_token":"` + setupToken + `"}`
+	if w := post(ok); w.Code != 200 || !strings.Contains(w.Body.String(), `"token"`) {
+		t.Fatalf("right setup token: status=%d want 200 with a token; body=%s", w.Code, w.Body.String())
+	}
+	if hash, _, err := m.pg.GetSetting(authPassKey); err != nil || hash == "" {
+		t.Fatalf("password not stored: err=%v", err)
+	}
+
+	// The token is spent: even with the password row removed, it is refused.
+	clearPassword()
+	if w := post(ok); w.Code != 403 {
+		t.Fatalf("reused setup token: status=%d want 403; body=%s", w.Code, w.Body.String())
+	}
+}
+
 func TestValidatePassword(t *testing.T) {
 	for _, tc := range []struct {
 		name, pw string

server/conversations.go

@@ -103,7 +103,7 @@ func (s *Server) pgCreateConversation(w http.ResponseWriter, r *http.Request) {
 	}
 	if req.LLMProfileID != nil {
 		if _, ok := s.loadProfileConfig(*req.LLMProfileID); !ok {
-			writeErr(w, 400, "The specified LLM profile does not exist or has no API Key set")
+			writeErr(w, 400, "The specified LLM profile does not exist or is not usable (no API key for the key-requiring formats)")
 			return
 		}
 	}
@@ -136,7 +136,7 @@ func (s *Server) pgUpdateConversation(w http.ResponseWriter, r *http.Request) {
 	}
 	if req.LLMProfileID != nil {
 		if _, ok := s.loadProfileConfig(*req.LLMProfileID); !ok {
-			writeErr(w, 400, "The specified LLM profile does not exist or has no API Key set")
+			writeErr(w, 400, "The specified LLM profile does not exist or is not usable (no API key for the key-requiring formats)")
 			return
 		}
 	}

server/env_scrub.go

@@ -0,0 +1,30 @@
+package server
+
+import "os"
+
+// secretEnv lists the credentials read from the environment only at startup:
+// ARTEX_PG_DSN by NewManager, the API keys by agent.FromEnv inside New.
+var secretEnv = []string{"ARTEX_PG_DSN", "ANTHROPIC_API_KEY", "OPENAI_API_KEY"}
+
+// ScrubSecretEnv removes secretEnv from the process environment once startup has
+// read it. Child processes (agent shells, custom tools, MCP stdio servers) copy
+// os.Environ() when they are spawned, so they no longer inherit these values.
+//
+// It runs inside New, after both readers and before the MCP discovery goroutine
+// and the task resume loop: both can spawn children, and resuming a task starts
+// workers that run shells. Scrubbing in main after New returned left that window
+// open.
+func ScrubSecretEnv() {
+	for _, k := range secretEnv {
+		_ = os.Unsetenv(k)
+	}
+}
+
+// Option configures New. Options keep the test call sites unchanged: only the
+// binary opts into scrubbing, because tests create several servers in one
+// process and still need ARTEX_PG_DSN on later calls.
+type Option func(*Server)
+
+// WithSecretScrub makes New drop the startup secrets after reading them, before
+// the MCP discovery goroutine and the task resume loop can spawn children.
+func WithSecretScrub() Option { return func(s *Server) { s.scrubSecrets = true } }

server/env_scrub_test.go

@@ -0,0 +1,36 @@
+package server
+
+import (
+	"os/exec"
+	"strings"
+	"testing"
+)
+
+// A child process spawned after startup must not see the startup secrets, while
+// other variables still pass through.
+func TestScrubSecretEnvHidesSecretsFromChildren(t *testing.T) {
+	sh, err := exec.LookPath("sh")
+	if err != nil {
+		t.Skip("no sh in PATH")
+	}
+	for _, k := range secretEnv {
+		t.Setenv(k, "leak-"+k)
+	}
+	t.Setenv("ARTEX_SCRUB_KEEP", "kept")
+
+	ScrubSecretEnv()
+
+	out, err := exec.Command(sh, "-c", "env").Output()
+	if err != nil {
+		t.Fatal(err)
+	}
+	env := "\n" + string(out)
+	for _, k := range secretEnv {
+		if strings.Contains(env, "\n"+k+"=") {
+			t.Errorf("%s still reaches child processes", k)
+		}
+	}
+	if !strings.Contains(env, "\nARTEX_SCRUB_KEEP=kept\n") {
+		t.Errorf("non-secret variable did not reach the child process")
+	}
+}

server/llm_ollama_test.go

@@ -0,0 +1,272 @@
+package server
+
+import (
+	"bytes"
+	"context"
+	"encoding/json"
+	"net/http"
+	"net/http/httptest"
+	"sync/atomic"
+	"testing"
+	"time"
+
+	"github.com/Autumn-27/artex/agent"
+	"github.com/Autumn-27/artex/db"
+	"github.com/Autumn-27/norma/llm"
+)
+
+// ConfigFrom("ollama", ...) must normalize to the keyless sentinel with our own
+// defaults, never the SDK's OpenAI-public-endpoint fallback.
+func TestOllamaConfigFromKeylessDefaults(t *testing.T) {
+	c := agent.ConfigFrom("ollama", "", "", "", "")
+	if c.Format != agent.FormatOllama {
+		t.Fatalf("format=%v, want FormatOllama", c.Format)
+	}
+	if c.Provider() != "ollama" {
+		t.Fatalf("Provider()=%q", c.Provider())
+	}
+	if c.BaseURL != agent.DefaultOllamaBaseURL {
+		t.Fatalf("BaseURL=%q, want default %q", c.BaseURL, agent.DefaultOllamaBaseURL)
+	}
+	if c.Model == "" {
+		t.Fatal("model must default to a non-empty value")
+	}
+	if c.NeedsAPIKey() {
+		t.Fatal("ollama must not require an API key")
+	}
+	// A full endpoint URL is tolerated (the provider appends /chat/completions).
+	e := agent.ConfigFrom("ollama", "llama3.2", "http://box:11434/v1/chat/completions", "", "")
+	if e.BaseURL != "http://box:11434/v1" {
+		t.Fatalf("full endpoint URL not normalized: %q", e.BaseURL)
+	}
+	// A custom key is preserved (e.g. an authenticated Ollama gateway).
+	k := agent.ConfigFrom("ollama", "m", "http://box:11434", "ollama-secret", "")
+	if k.APIKey != "ollama-secret" {
+		t.Fatalf("custom key lost: %q", k.APIKey)
+	}
+	// The natural Ollama URL has no /v1, but the provider appends
+	// /chat/completions, so the base must be normalized to the /v1 root.
+	n := agent.ConfigFrom("ollama", "m", "http://box:11434", "", "")
+	if n.BaseURL != "http://box:11434/v1" {
+		t.Fatalf("bare host URL not normalized: %q", n.BaseURL)
+	}
+}
+
+// The provider built from an Ollama config speaks OpenAI Chat Completions on
+// the wire (the SDK appends /chat/completions) with the placeholder Bearer (no
+// real key was configured, no OPENAI_API_KEY env leak).
+func TestOllamaProviderWiresOpenAIProtocol(t *testing.T) {
+	var gotPath, gotAuth, gotModel atomic.Value
+	fake := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		gotPath.Store(r.URL.Path)
+		gotAuth.Store(r.Header.Get("Authorization"))
+		var body struct {
+			Model string `json:"model"`
+		}
+		_ = json.NewDecoder(r.Body).Decode(&body)
+		gotModel.Store(body.Model)
+		w.Header().Set("Content-Type", "application/json")
+		_, _ = w.Write([]byte(`{"id":"1","object":"chat.completion","model":"llama3.2","choices":[{"index":0,"message":{"role":"assistant","content":"OK"},"finish_reason":"stop"}],"usage":{"prompt_tokens":1,"completion_tokens":1}}`))
+	}))
+	t.Cleanup(fake.Close)
+
+	cfg := agent.ConfigFrom("ollama", "llama3.2", fake.URL, "", "")
+	cfg.Stream = false // non-streaming: the fake answers one plain JSON body
+	prov, err := cfg.NewProvider()
+	if err != nil {
+		t.Fatal(err)
+	}
+	ctx, cancel := context.WithTimeout(context.Background(), 10*time.Second)
+	defer cancel()
+	req := llm.CompletionRequest{
+		System:   []string{"You are a connection test. Output OK."},
+		Messages: []llm.Message{llm.UserText("ping")},
+	}
+	if _, _, _, err = prov.Complete(ctx, req); err != nil {
+		t.Fatalf("Complete: %v", err)
+	}
+	// ConfigFrom normalizes the fake server URL (no /v1) onto the OpenAI root,
+	// so the request must land on /v1/chat/completions. A path of /chat/completions
+	// would mean the base URL leaked through unnormalized (or fell back to OpenAI).
+	if p, _ := gotPath.Load().(string); p != "/v1/chat/completions" {
+		t.Fatalf("path=%q, want /v1/chat/completions", p)
+	}
+	if a, _ := gotAuth.Load().(string); a != "Bearer ollama" {
+		t.Fatalf("Authorization=%q, want placeholder Bearer ollama", a)
+	}
+	if m, _ := gotModel.Load().(string); m != "llama3.2" {
+		t.Fatalf("model on wire=%q", m)
+	}
+}
+
+// A keyless Ollama profile must be usable end to end: stored round-trips,
+// loadProfileConfig accepts it, the task chain validates it, the failover pool
+// includes it, and the connection test completes against a fake Ollama.
+func TestOllamaKeylessProfileLifecycle(t *testing.T) {
+	m, err := NewManager(t.TempDir(), "")
+	if err != nil {
+		t.Fatal(err)
+	}
+	t.Cleanup(func() { m.Close() })
+	if m.pg == nil {
+		t.Skip("no postgres for this run")
+	}
+	ctx, cancel := context.WithCancel(context.Background())
+	cancel()
+	s := New(ctx, m, t.TempDir(), t.TempDir(), t.TempDir())
+
+	fake := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		w.Header().Set("Content-Type", "application/json")
+		_, _ = w.Write([]byte(`{"id":"1","object":"chat.completion","model":"llama3.2","choices":[{"index":0,"message":{"role":"assistant","content":"OK"},"finish_reason":"stop"}],"usage":{"prompt_tokens":1,"completion_tokens":1}}`))
+	}))
+	t.Cleanup(fake.Close)
+
+	pid, err := m.pg.SaveProfile(&db.LLMProfile{
+		Name: "ollama-local", Format: "ollama", Model: "llama3.2",
+		BaseURL: fake.URL, APIKey: "", IsDefault: true,
+	})
+	if err != nil {
+		t.Fatal(err)
+	}
+	t.Cleanup(func() { _, _ = m.pg.Exec("DELETE FROM llm_profiles WHERE id=$1", pid) })
+
+	stored, err := m.pg.ProfileByID(pid)
+	if err != nil || stored == nil {
+		t.Fatalf("ProfileByID: %v %v", stored, err)
+	}
+
+	if _, ok := s.loadProfileConfig(pid); !ok {
+		t.Fatal("keyless ollama profile must load")
+	}
+	if err := s.validateTaskProfileIDs([]int64{pid}); err != nil {
+		t.Fatalf("task chain must accept keyless ollama: %v", err)
+	}
+	pool, err := m.pg.PoolProfiles()
+	if err != nil {
+		t.Fatal(err)
+	}
+	found := false
+	for _, p := range pool {
+		if p.ID == pid {
+			found = true
+		}
+	}
+	if !found {
+		t.Fatal("keyless ollama profile must join the failover pool")
+	}
+
+	// Connection test: no api_key in the form either.
+	body, _ := json.Marshal(map[string]any{
+		"profile_id": pid, "provider": "ollama", "model": "llama3.2",
+		"base_url": fake.URL, "api_key": "", "streaming": false,
+	})
+	req := httptest.NewRequest(http.MethodPost, "/api/llm/test", bytes.NewReader(body))
+	req.Header.Set("Content-Type", "application/json")
+	rec := httptest.NewRecorder()
+	s.testLLM(rec, req)
+	if rec.Code != http.StatusOK {
+		t.Fatalf("handler status %d", rec.Code)
+	}
+	var resp struct {
+		OK    bool   `json:"ok"`
+		Error string `json:"error"`
+	}
+	_ = json.Unmarshal(rec.Body.Bytes(), &resp)
+	if !resp.OK {
+		t.Fatalf("connection test on keyless ollama failed: %s", resp.Error)
+	}
+}
+
+// Model listing for an Ollama server works without an API key and tolerates a
+// base URL with or without the /v1 suffix.
+func TestOllamaListModels(t *testing.T) {
+	var calls atomic.Value // []string paths
+	fake := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		w.Header().Set("Content-Type", "application/json")
+		paths, _ := calls.Load().([]string)
+		calls.Store(append(paths, r.URL.Path))
+		_, _ = w.Write([]byte(`{"object":"list","data":[{"id":"llama3.2"},{"id":"qwen2.5:14b"}]}`))
+	}))
+	t.Cleanup(fake.Close)
+
+	m, err := NewManager(t.TempDir(), "")
+	if err != nil {
+		t.Fatal(err)
+	}
+	t.Cleanup(func() { m.Close() })
+	if m.pg == nil {
+		t.Skip("no postgres for this run")
+	}
+	ctx, cancel := context.WithCancel(context.Background())
+	cancel()
+	s := New(ctx, m, t.TempDir(), t.TempDir(), t.TempDir())
+
+	call := func(baseURL string) (bool, []string) {
+		body, _ := json.Marshal(map[string]any{"provider": "ollama", "base_url": baseURL, "api_key": ""})
+		req := httptest.NewRequest(http.MethodPost, "/api/llm/models", bytes.NewReader(body))
+		req.Header.Set("Content-Type", "application/json")
+		rec := httptest.NewRecorder()
+		s.pgListModels(rec, req)
+		var resp struct {
+			OK     bool     `json:"ok"`
+			Models []string `json:"models"`
+			Error  string   `json:"error"`
+		}
+		_ = json.Unmarshal(rec.Body.Bytes(), &resp)
+		return resp.OK, resp.Models
+	}
+	// Base WITH /v1.
+	ok, models := call(fake.URL + "/v1")
+	if !ok || len(models) != 2 {
+		t.Fatalf("listing with /v1 base failed: ok=%v models=%v", ok, models)
+	}
+	// Base WITHOUT /v1 must still be normalized onto the OpenAI-compatible path
+	// (/v1/models); both calls hit the same endpoint.
+	ok2, models2 := call(fake.URL)
+	if !ok2 || len(models2) != 2 {
+		t.Fatalf("listing without /v1 base failed: ok=%v models=%v", ok2, models2)
+	}
+	if paths, _ := calls.Load().([]string); len(paths) != 2 || paths[0] != "/v1/models" || paths[1] != "/v1/models" {
+		t.Fatalf("both calls must hit /v1/models, got %v", paths)
+	}
+}
+
+// The environment path must normalize the base URL too: ARTEX_LLM_BASE_URL is
+// the natural way to configure a LAN Ollama server, and it used to send chat to
+// /chat/completions instead of /v1/chat/completions.
+func TestOllamaFromEnvNormalizesBaseURL(t *testing.T) {
+	t.Setenv("ARTEX_LLM_PROVIDER", "ollama")
+	t.Setenv("OPENAI_API_KEY", "")
+	t.Setenv("ARTEX_LLM_BASE_URL", "http://box:11434")
+	c, ok := agent.FromEnv()
+	if !ok {
+		t.Fatal("ollama env config must be usable")
+	}
+	if c.BaseURL != "http://box:11434/v1" {
+		t.Fatalf("env base URL not normalized: %q", c.BaseURL)
+	}
+	t.Setenv("ARTEX_LLM_BASE_URL", "http://box:11434/v1/chat/completions")
+	c, ok = agent.FromEnv()
+	if !ok {
+		t.Fatal("ollama env config must be usable")
+	}
+	if c.BaseURL != "http://box:11434/v1" {
+		t.Fatalf("full endpoint URL not normalized: %q", c.BaseURL)
+	}
+}
+
+// An Ollama config resolved from the environment must not inherit OPENAI_API_KEY:
+// Ollama ignores the bearer token, so inheritance would only ship the cloud key
+// to the local or LAN endpoint.
+func TestOllamaFromEnvDoesNotInheritOpenAIKey(t *testing.T) {
+	t.Setenv("ARTEX_LLM_PROVIDER", "ollama")
+	t.Setenv("OPENAI_API_KEY", "sk-cloud-secret")
+	t.Setenv("ARTEX_LLM_BASE_URL", "")
+	c, ok := agent.FromEnv()
+	if !ok {
+		t.Fatal("ollama env config must be usable without a key")
+	}
+	if c.APIKey != "" {
+		t.Fatalf("ollama inherited a cloud key: %q", c.APIKey)
+	}
+}

server/orchestration.go

@@ -293,7 +293,7 @@ func (s *Server) toolSpawnTask() actool.CoreTool {
 			var pin *int64
 			if id := parseProfileID(a.LLMProfileID); id > 0 {
 				if _, ok := s.loadProfileConfig(id); !ok {
-					return actool.Errorf(fmt.Sprintf("LLM profile #%d not found or has no API key set", id)), nil
+					return actool.Errorf(fmt.Sprintf("LLM profile #%d not found or not usable (no API key for the key-requiring formats)", id)), nil
 				}
 				pin = &id
 			} else if a.ParentRef != "" {

server/server.go

@@ -46,6 +46,11 @@ type Server struct {
 	skillDir string // root directory for skill subdirectories
 	jwtKey   []byte // HS256 signing key loaded from / generated into dataDir/jwt.key
 
+	scrubSecrets bool // drop ARTEX_PG_DSN and the API keys from the environment once read (binary only)
+
+	setupMu    sync.Mutex
+	setupToken string // one-time token POST /api/auth/init requires; empty when none is active
+
 	// 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.
@@ -140,7 +145,7 @@ type triggeredRun struct {
 	mergeable bool   // true for finding/goal event triggers (merge by taskID)
 }
 
-func New(ctx context.Context, m *Manager, skillDir string, dataDir string, keyDir string) *Server {
+func New(ctx context.Context, m *Manager, skillDir string, dataDir string, keyDir string, opts ...Option) *Server {
 	key, err := loadOrCreateJWTKey(keyDir, dataDir)
 	if err != nil {
 		log.Fatalf("[auth] JWT key: %v", err)
@@ -151,6 +156,7 @@ func New(ctx context.Context, m *Manager, skillDir string, dataDir string, keyDi
 		profChatAgents: map[int64]*agent.ChatAgent{},
 		provByProfile:  map[int64]*provEntry{}, llmHealth: newLLMHealthRegistry(m.pg),
 		taskAgents: map[string]*taskAgentBundle{}, archiveWake: make(chan struct{}, 1)}
+	s.initSetupToken()
 	s.initSideQuestions()
 	// Circuit-breaker thresholds/cooldowns are hot parameters on the failure path; push the global retry policy to the Registry once at startup,
 	// and again on every policy save (saveLLMRetryPolicy).
@@ -232,9 +238,6 @@ func New(ctx context.Context, m *Manager, skillDir string, dataDir string, keyDi
 		// Vulnerability IM push delivery engine. Sits beside the Scheduler but independent: push latency demands (3s)
 		// differ from the triggers' business cadence, and neither's failures drag the other down — a stuck push must not affect agent triggers.
 		go newNotifier(s).Run(s.ctx)
-		// Fill the tool cache for any enabled MCP that has none yet (notably the
-		// seeded browser MCP on first run). Async so it never blocks startup.
-		go s.discoverEmptyMCPsOnStartup()
 		logSink.SetDB(ctx, m.pg) // restore last 100 log rows and enable async persistence
 	}
 	// precedence: persisted DB config > env.
@@ -253,6 +256,16 @@ func New(ctx context.Context, m *Manager, skillDir string, dataDir string, keyDi
 	} else {
 		log.Printf("[engine] no LLM provider configured — engine idle until set via /api/llm or env")
 	}
+	// Both env readers have now run (NewManager for the DSN, FromEnv above for the
+	// keys). Drop the secrets before anything below can spawn a child process:
+	// restoring runtimes resumes workers that run shells, and MCP discovery starts
+	// stdio servers.
+	if s.scrubSecrets {
+		ScrubSecretEnv()
+	}
+	// Fill the tool cache for any enabled MCP that has none yet (notably the
+	// seeded browser MCP on first run). Async so it never blocks startup.
+	go s.discoverEmptyMCPsOnStartup()
 	s.restoreTaskRuntimes()
 	go s.reconcileConcurrency()
 	s.startTaskArchiveWorker()
@@ -335,7 +348,8 @@ func (s *Server) loadLLMConfig() (agent.Config, bool) {
 	cfg.MaxTokens, cfg.MaxTokensField = p.MaxTokens, p.MaxTokensField
 	cfg.SessionHeaderKey = p.SessionHeaderKey
 	s.applyProfileRetry(&cfg, p)
-	if cfg.APIKey == "" {
+	// Ollama profiles run keyless (local server); the other formats need a key.
+	if cfg.APIKey == "" && cfg.NeedsAPIKey() {
 		return cfg, false
 	}
 	s.cfgMu.Lock()
@@ -491,7 +505,8 @@ func (s *Server) applyLLM(cfg agent.Config) error {
 }
 
 // loadProfileConfig builds an agent.Config from a specific profile id (with its key).
-// ok=false when the profile is missing or has no api key.
+// ok=false when the profile is missing or unusable (no api key for the key-requiring
+// formats).
 func (s *Server) loadProfileConfig(id int64) (agent.Config, bool) {
 	p, err := s.m.pg.ProfileByID(id)
 	if err != nil || p == nil {
@@ -506,7 +521,7 @@ func (s *Server) loadProfileConfig(id int64) (agent.Config, bool) {
 	cfg.MaxTokens, cfg.MaxTokensField = p.MaxTokens, p.MaxTokensField
 	cfg.SessionHeaderKey = p.SessionHeaderKey
 	s.applyProfileRetry(&cfg, p)
-	if cfg.APIKey == "" {
+	if cfg.APIKey == "" && cfg.NeedsAPIKey() {
 		return cfg, false
 	}
 	return cfg, true
@@ -1343,12 +1358,13 @@ func (s *Server) setLLM(w http.ResponseWriter, r *http.Request) {
 		}
 		cfg.ContextWindowK = k
 	}
-	if cfg.APIKey == "" {
+	// Keyless formats (Ollama) need no key; the others keep/require one.
+	if cfg.APIKey == "" && cfg.NeedsAPIKey() {
 		s.cfgMu.Lock()
 		cfg.APIKey = s.llmCfg.APIKey // keep existing key if not re-entered
 		s.cfgMu.Unlock()
 	}
-	if cfg.APIKey == "" {
+	if cfg.APIKey == "" && cfg.NeedsAPIKey() {
 		writeErr(w, 400, "api_key required")
 		return
 	}
@@ -1441,28 +1457,30 @@ func (s *Server) testLLM(w http.ResponseWriter, r *http.Request) {
 	// see TestConnection). Endpoints like opencode zen that mandate x-opencode-session return
 	// 400 without it, so the test path must send it too — otherwise "chat works, test 400s".
 	cfg.SessionHeaderKey = req.SessionHeaderKey
-	// API key resolution precedence: form input > the named profile's stored key > the global config's key.
-	// A stored profile's key is never sent back to the browser, so the form is empty when testing a stored profile and the key comes from the DB.
-	// Same for the session header name: when the form omits it, the stored profile's value is the fallback.
-	if stored != nil && (cfg.APIKey == "" || cfg.SessionHeaderKey == "") {
-		// Trim like ConfigFrom does for form values: a key stored with
-		// surrounding whitespace runs fine in conversations (the runtime
-		// path re-trims) but would be sent raw here and fail auth.
-		if cfg.APIKey == "" {
+	// The custom session header name follows that profile regardless of format:
+	// when the form omits it, the stored profile's value is the fallback.
+	if stored != nil && cfg.SessionHeaderKey == "" {
+		cfg.SessionHeaderKey = strings.TrimSpace(stored.SessionHeaderKey)
+	}
+	// Keyless formats (Ollama) skip the key resolution entirely.
+	if cfg.NeedsAPIKey() {
+		// API key resolution precedence: form input > the named profile's stored key > the global config's key.
+		// A stored profile's key is never sent back to the browser, so the form is empty when testing a stored profile and the key comes from the DB.
+		if stored != nil && cfg.APIKey == "" {
+			// Trim like ConfigFrom does for form values: a key stored with
+			// surrounding whitespace runs fine in conversations (the runtime
+			// path re-trims) but would be sent raw here and fail auth.
 			cfg.APIKey = strings.TrimSpace(stored.APIKey)
 		}
-		if cfg.SessionHeaderKey == "" {
-			cfg.SessionHeaderKey = strings.TrimSpace(stored.SessionHeaderKey)
+		if cfg.APIKey == "" {
+			s.cfgMu.Lock()
+			cfg.APIKey = s.llmCfg.APIKey
+			s.cfgMu.Unlock()
+		}
+		if cfg.APIKey == "" {
+			writeJSON(w, 200, map[string]any{"ok": false, "error": "no API key provided"})
+			return
 		}
-	}
-	if cfg.APIKey == "" {
-		s.cfgMu.Lock()
-		cfg.APIKey = s.llmCfg.APIKey
-		s.cfgMu.Unlock()
-	}
-	if cfg.APIKey == "" {
-		writeJSON(w, 200, map[string]any{"ok": false, "error": "no API key provided"})
-		return
 	}
 	// Retry params do [not] go into the connection test: the test has a 30s hard timeout, and stacking the configured retry counts/long intervals on top
 	// would only turn a perfectly usable endpoint into a "timeout failure". The test asks "does this endpoint work at all"; retry pacing
@@ -1621,7 +1639,7 @@ func (s *Server) validateTaskProfileIDs(ids []int64) error {
 		}
 		seen[id] = true
 		if _, ok := s.loadProfileConfig(id); !ok {
-			return fmt.Errorf("LLM profile #%d not found or has no API key set", id)
+			return fmt.Errorf("LLM profile #%d not found or not usable (no API key for the key-requiring formats)", id)
 		}
 	}
 	return nil

server/server_mgmt.go

@@ -2037,12 +2037,13 @@ func (s *Server) pgLLMPoolReset(w http.ResponseWriter, r *http.Request) {
 }
 
 // pgListModels fetches available models from the provider's API endpoint.
-// Supports OpenAI-format (GET /models) and Anthropic-format (GET /v1/models); for
+// Supports OpenAI-format (GET /models), Anthropic-format (GET /v1/models) and
+// Ollama (keyless OpenAI-compatible GET /v1/models on the local server); for
 // Anthropic-compatible third parties (e.g. DeepSeek) whose model list lives only on
 // the OpenAI path, it falls back to the OpenAI endpoint at the stripped root.
 func (s *Server) pgListModels(w http.ResponseWriter, r *http.Request) {
 	var req struct {
-		Provider  string `json:"provider"` // "openai" | "anthropic"
+		Provider  string `json:"provider"` // "openai" | "anthropic" | "ollama"
 		BaseURL   string `json:"base_url"`
 		APIKey    string `json:"api_key"`
 		Proxy     string `json:"proxy"`
@@ -2059,13 +2060,14 @@ func (s *Server) pgListModels(w http.ResponseWriter, r *http.Request) {
 			apiKey = p.APIKey
 		}
 	}
-	if apiKey == "" {
+	provider := strings.TrimSpace(req.Provider)
+	// Ollama serves its model list keyless on the local server: no key gate.
+	if apiKey == "" && provider != "ollama" {
 		writeJSON(w, 200, map[string]any{"ok": false, "error": "no API Key provided"})
 		return
 	}
 
 	baseURL := strings.TrimRight(strings.TrimSpace(req.BaseURL), "/")
-	provider := strings.TrimSpace(req.Provider)
 
 	// Candidate endpoints to try in order. Some Anthropic-compatible providers
 	// (e.g. DeepSeek) implement /v1/messages under an /anthropic path but expose
@@ -2097,6 +2099,22 @@ func (s *Server) pgListModels(w http.ResponseWriter, r *http.Request) {
 		}
 		b := strings.TrimRight(strings.TrimSuffix(strings.TrimSuffix(baseURL, "/chat/completions"), "/responses"), "/")
 		candidates = append(candidates, candidate{b + "/models", bearerHdr()})
+	case "ollama":
+		// OpenAI-compatible list at {base}/v1/models. Tolerate base URLs with or
+		// without the /v1 suffix (and a full endpoint URL). No auth needed; when a
+		// key was provided anyway, send it as Bearer (Ollama ignores it).
+		if baseURL == "" {
+			baseURL = agent.DefaultOllamaBaseURL
+		}
+		b := strings.TrimSuffix(baseURL, "/chat/completions")
+		if !strings.HasSuffix(b, "/v1") {
+			b += "/v1"
+		}
+		hdr := http.Header{}
+		if apiKey != "" {
+			hdr.Set("Authorization", "Bearer "+apiKey)
+		}
+		candidates = append(candidates, candidate{b + "/models", hdr})
 	default: // anthropic
 		if baseURL == "" {
 			baseURL = "https://api.anthropic.com"

server/task_resolution.go

@@ -29,7 +29,8 @@ func (s *Server) resolutionFromProfile(p *db.LLMProfile, source string) taskLLMR
 		Model:     p.Model,
 		Source:    source,
 	}
-	if p.APIKey == "" {
+	// Ollama profiles are keyless by design; the other formats need a key.
+	if p.APIKey == "" && p.Format != "ollama" {
 		result.Reason = "LLM profile has no API Key set"
 		return result
 	}

update.sh

@@ -1,5 +1,5 @@
 #!/usr/bin/env bash
-# ARTEX update script: (1) Docker update (pull new image and rebuild)  (2) local build update (rebuild the binary)
+# ARTEX update script: (1) Docker update (rebuild the image from this checkout)  (2) local build update (rebuild the binary)
 # Counterpart of install.sh: install handles the first-time setup, update upgrades to a new version.
 # No manual DB migration needed — artex idempotently re-runs schema.sql on every start (with ADD COLUMN/CREATE
 # INDEX IF NOT EXISTS), so "restart equals migrate". Data (pgdata volume, ./data, ./skills) is not affected.
@@ -17,7 +17,16 @@ sync_repo(){
   [ -d .git ] && command -v git >/dev/null 2>&1 || { warn "not a git working copy, skipping git pull"; return; }
   [ "$(ask 'Pull the latest code (git pull --ff-only)? (y/n)' y)" = y ] || return
   if ! git pull --ff-only; then
-    warn "git pull could not fast-forward (local changes or diverged branches) — resolve manually and retry; continuing with the current code"
+    # Continuing would rebuild the OLD tree and still print "Build complete":
+    # exactly the silent stale upgrade this script exists to prevent.
+    cat >&2 <<'RECOVER'
+[!] git pull could not fast-forward. This happens when the checkout has local
+    changes, or when it was cloned before the history rewrite of 2026-10-23.
+    Refusing to rebuild stale code. Recover, then rerun this script:
+      keep your changes:  git stash && git fetch origin && git reset --hard origin/main && git stash pop
+      discard them:       git fetch origin && git reset --hard origin/main
+RECOVER
+    die "aborting: nothing was rebuilt"
   fi
 }
 
@@ -27,23 +36,12 @@ update_docker(){
     || die "docker / docker compose not detected; install and deploy with ./install.sh first"
   [ -f .env ] || die ".env not found; run ./install.sh first to complete the initial deployment"
 
-  # Optional: upgrade to a specific version tag (empty = keep ARTEX_TAG from .env, default latest)
-  local tag; tag="$(ask 'Target image tag (press Enter to keep .env / latest)' '')"
-  if [ -n "$tag" ]; then
-    if grep -q '^ARTEX_TAG=' .env; then
-      sed -i.bak "s|^ARTEX_TAG=.*|ARTEX_TAG=${tag}|" .env && rm -f .env.bak
-    else
-      printf '\nARTEX_TAG=%s\n' "$tag" >> .env
-    fi
-    ok "ARTEX_TAG set to ${tag}"
-  fi
-
   # Touch only artex: postgres is pinned to 16-alpine and does not need upgrading along
   # (pulling it wastes bandwidth, and a major version bump risks compatibility). artex
   # declares depends_on postgres, so an up with the service name starts pg if needed and leaves a running one untouched.
-  info "Pulling new image (artex only)…"
-  docker compose pull artex
-  info "Rebuilding and starting (schema migrates automatically on artex restart)…"
+  info "Building the image from this checkout (artex only)…"
+  docker compose build artex
+  info "Recreating artex (schema migrates automatically on restart)…"
   docker compose up -d artex
   ok "Update complete → http://localhost:8787"
   info "View logs: docker compose logs -f artex"

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

@@ -14,6 +14,7 @@ export default function SetupPage() {
   const router = useRouter();
   const [password, setPassword] = useState("");
   const [confirm, setConfirm] = useState("");
+  const [setupToken, setSetupToken] = useState("");
   const [error, setError] = useState("");
   const [loading, setLoading] = useState(false);
   const [checking, setChecking] = useState(true);
@@ -47,7 +48,7 @@ export default function SetupPage() {
     setLoading(true);
     setError("");
     try {
-      const { token } = await api.initPassword(password);
+      const { token } = await api.initPassword(password, setupToken.trim());
       auth.setToken(token);
       router.replace("/function/tasks");
     } catch (err) {
@@ -80,7 +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? 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 8 characters)"}
             </p>
           </div>
           {unavailable ? (
@@ -92,6 +93,17 @@ export default function SetupPage() {
             </div>
           ) : (
             <form onSubmit={handleSubmit} className="flex flex-col gap-4">
+              <div className="space-y-1.5">
+                <Label htmlFor="setup-token">Setup token</Label>
+                <Input
+                  id="setup-token"
+                  value={setupToken}
+                  onChange={(e) => setSetupToken(e.target.value)}
+                  placeholder="Printed in the server log at startup"
+                  autoFocus
+                  autoComplete="off"
+                />
+              </div>
               <div className="space-y-1.5">
                 <Label htmlFor="password">New password</Label>
                 <Input
@@ -100,7 +112,6 @@ export default function SetupPage() {
                   value={password}
                   onChange={(e) => setPassword(e.target.value)}
                   placeholder="At least 8 characters"
-                  autoFocus
                   autoComplete="new-password"
                 />
               </div>
@@ -116,7 +127,11 @@ export default function SetupPage() {
                 />
               </div>
               {error && <p className="text-sm text-destructive">{error}</p>}
-              <Button type="submit" className="w-full" disabled={loading || !password || !confirm}>
+              <Button
+                type="submit"
+                className="w-full"
+                disabled={loading || !setupToken.trim() || !password || !confirm}
+              >
                 {loading ? "Saving..." : "Set password and sign in"}
               </Button>
             </form>

web/src/app/(main)/system/llm/page.tsx

@@ -49,13 +49,18 @@ const MAX_TOKENS_FIELDS: { value: string; label: string }[] = [
   { value: NONE, label: "max_tokens (default)" },
   { value: "max_completion_tokens", label: "max_completion_tokens" },
 ];
-// The other two formats hard-code their field names, so the option is meaningless for them; the hint texts spell this out.
+// The other formats hard-code their field names, so the option is meaningless for them; the hint texts spell this out.
 const MAX_TOKENS_FIELD_HINTS: Record<string, string> = {
   openai:
     "Which key carries the cap. max_tokens is the default and the vast majority of compatible gateways only accept it; OpenAI's official reasoning models (the o series / GPT-5) conversely only accept max_completion_tokens and fail outright with unsupported_parameter when they receive max_tokens.",
   anthropic: "Only selectable for the openai format. Anthropic's field name is fixed as max_tokens.",
   "openai-responses": "Only selectable for the openai format. The Responses API's field name is fixed as max_output_tokens.",
+  ollama: "Only selectable for the openai format. Ollama's field name is fixed as max_tokens.",
 };
+// LLM profile formats. The default in the create form is "anthropic"; "ollama" is the
+// keyless local-server format (base URL defaults to http://127.0.0.1:11434/v1).
+type ProfileFormat = "anthropic" | "openai" | "openai-responses" | "ollama";
+const OLLAMA_DEFAULT_BASE_URL = "http://127.0.0.1:11434/v1";
 const EFFORT_LEVELS: { value: string; label: string }[] = [
   { value: NONE, label: "Not sent (default)" },
   { value: "low", label: "low" },
@@ -72,10 +77,11 @@ function cooldownText(secs: number) {
 }
 
 // The "healthy or not" badge a profile shows on its card. A profile without a Key cannot send any request at all — say that before the breaker;
-// the other states come from the pool's breaker records (with polling off no new records appear, so "healthy" = no known failure).
+// the exception is Ollama, a keyless local server that IS callable without a key. The other states come from the pool's breaker
+// records (with polling off no new records appear, so "healthy" = no known failure).
 type Health = { label: string; cls: string; hint?: string };
 function healthOf(p: LLMProfile, m?: LLMPoolMember): Health {
-  if (!p.api_key_hint) {
+  if (!p.api_key_hint && p.format !== "ollama") {
     return {
       label: "No Key configured",
       cls: "border-muted-foreground/40 text-muted-foreground",
@@ -215,7 +221,8 @@ function PoolSheet({
                 </div>
                 {inChain.length < 2 && (
                   <p className="text-muted-foreground text-xs">
-                    Only {inChain.length} usable profile(s) right now, so the pool has no effect — at least 2 profiles with an API Key set and pool participation are required.
+                    Only {inChain.length} usable profile(s) right now, so the pool has no effect — at least 2 usable profiles (API Key set, or Ollama) with pool participation are
+                    required.
                   </p>
                 )}
                 {chain.map((m) => {
@@ -311,7 +318,7 @@ function ProfileSheet({
 }) {
   const isNew = !profile;
   const [name, setName] = React.useState("");
-  const [format, setFormat] = React.useState<"anthropic" | "openai" | "openai-responses">("anthropic");
+  const [format, setFormat] = React.useState<ProfileFormat>("anthropic");
   const [model, setModel] = React.useState("");
   const [baseUrl, setBaseUrl] = React.useState("");
   const [proxy, setProxy] = React.useState("");
@@ -340,7 +347,11 @@ function ProfileSheet({
   React.useEffect(() => {
     if (!open) return;
     setName(profile?.name ?? "");
-    setFormat(profile?.format === "openai" || profile?.format === "openai-responses" ? profile.format : "anthropic");
+    setFormat(
+      profile?.format === "openai" || profile?.format === "openai-responses" || profile?.format === "ollama"
+        ? profile.format
+        : "anthropic",
+    );
     setModel(profile?.model ?? "");
     setBaseUrl(profile?.base_url ?? "");
     setProxy(profile?.proxy ?? "");
@@ -457,6 +468,13 @@ function ProfileSheet({
     }
   }
 
+  let keyPlaceholder = "sk-…";
+  if (format === "ollama") {
+    keyPlaceholder = "optional — a local Ollama server runs keyless";
+  } else if (keyHint) {
+    keyPlaceholder = `Set (${keyHint}); leave empty to keep it`;
+  }
+
   return (
     <Sheet open={open} onOpenChange={onOpenChange}>
       <SheetContent
@@ -492,7 +510,7 @@ function ProfileSheet({
             </div>
             <div className="grid gap-2">
               <Label>Format</Label>
-              <Select value={format} onValueChange={(v) => setFormat(v as "anthropic" | "openai" | "openai-responses")}>
+              <Select value={format} onValueChange={(v) => setFormat(v as ProfileFormat)}>
                 <SelectTrigger>
                   <SelectValue placeholder="Select a format" />
                 </SelectTrigger>
@@ -500,6 +518,7 @@ function ProfileSheet({
                   <SelectItem value="anthropic">Anthropic</SelectItem>
                   <SelectItem value="openai">OpenAI (Chat Completions)</SelectItem>
                   <SelectItem value="openai-responses">OpenAI (Responses API)</SelectItem>
+                  <SelectItem value="ollama">Ollama (local, no key needed)</SelectItem>
                 </SelectContent>
               </Select>
             </div>
@@ -511,7 +530,7 @@ function ProfileSheet({
               <Input
                 id="p-model"
                 className="font-mono"
-                placeholder="claude-opus-4-8"
+                placeholder={format === "ollama" ? "llama3.2" : "claude-opus-4-8"}
                 value={model}
                 onChange={(e) => setModel(e.target.value)}
               />
@@ -557,7 +576,7 @@ function ProfileSheet({
             <Input
               id="p-base-url"
               className="font-mono"
-              placeholder="https://api.openai.com/v1"
+              placeholder={format === "ollama" ? OLLAMA_DEFAULT_BASE_URL : "https://api.openai.com/v1"}
               value={baseUrl}
               onChange={(e) => setBaseUrl(e.target.value)}
             />
@@ -595,11 +614,13 @@ function ProfileSheet({
           </div>
 
           <div className="grid gap-2">
-            <Label htmlFor="p-api-key">API Key</Label>
+            <Label htmlFor="p-api-key">
+              {format === "ollama" ? "API Key (optional for Ollama)" : "API Key"}
+            </Label>
             <Input
               id="p-api-key"
               type="password"
-              placeholder={keyHint ? `Set (${keyHint}); leave empty to keep it` : "sk-…"}
+              placeholder={keyPlaceholder}
               value={apiKey}
               onChange={(e) => setApiKey(e.target.value)}
             />

web/src/lib/api.ts

@@ -240,7 +240,8 @@ export const api = {
   // ---- auth ----
   authStatus: () => get<{ initialized: boolean }>("/auth/status"),
   login: (username: string, password: string) => post<{ token: string }>("/auth/login", { username, password }),
-  initPassword: (password: string) => post<{ token: string }>("/auth/init", { password }),
+  initPassword: (password: string, setupToken: string) =>
+    post<{ token: string }>("/auth/init", { password, setup_token: setupToken }),
   changePassword: (oldPassword: string, newPassword: string) =>
     post<{ ok: boolean }>("/auth/change-password", { old_password: oldPassword, new_password: newPassword }),
 

web/src/lib/types.ts

@@ -1028,7 +1028,7 @@ export interface NotificationDelivery {
 export interface LLMProfile {
   id: string;
   name: string;
-  format: "openai" | "anthropic" | "openai-responses";
+  format: "openai" | "anthropic" | "openai-responses" | "ollama";
   base_url?: string;
   proxy?: string;
   model: string;