diff --git a/.env.example b/.env.example
new file mode 100644
index 0000000..2252caf
--- /dev/null
+++ b/.env.example
@@ -0,0 +1,21 @@
+# Gateway RAR AUTOPASS — variabile de mediu (copiaza in .env; .env NU se comite).
+# Compose citeste .env automat. Prefix AUTOPASS_ pentru toate.
+
+# --- CRITIC: cheie criptare creds RAR (Fernet) ---
+# PARTAJATA intre api si worker (API cripteaza, worker decripteaza). Genereaza:
+#   python -c "from cryptography.fernet import Fernet; print(Fernet.generate_key().decode())"
+AUTOPASS_CREDS_KEY=
+
+# --- Auth API-key ---
+# true = orice /v1/* cere cheie valida (prod). false = dev (fara cheie -> cont id=1).
+AUTOPASS_REQUIRE_API_KEY=false
+
+# --- Worker ---
+# Send catre RAR. false = nu trimite (default, sigur pentru probe). true = end-to-end.
+AUTOPASS_WORKER_SEND_ENABLED=false
+# Dev: foloseste creds <test> din settings.xml cand submission-ul nu are creds criptate.
+AUTOPASS_WORKER_USE_TEST_CREDS=false
+
+# --- RAR ---
+# test | prod
+AUTOPASS_RAR_ENV=test
diff --git a/.gitignore b/.gitignore
index ffc24eb..9a8e21c 100644
--- a/.gitignore
+++ b/.gitignore
@@ -13,6 +13,7 @@ settings.xml
 *.key
 .env
 .env.*
+!.env.example
 
 # --- VFP: programe compilate (se regenerează din .prg) ---
 *.fxp
diff --git a/app/worker/healthcheck.py b/app/worker/healthcheck.py
new file mode 100644
index 0000000..f65a6bb
--- /dev/null
+++ b/app/worker/healthcheck.py
@@ -0,0 +1,48 @@
+"""Liveness probe pentru worker (T6) — folosit de healthcheck-ul Docker.
+
+Worker-ul nu e server HTTP, deci `restart: always` prinde doar procesul MORT,
+nu si worker-ul AGATAT (proces viu care nu mai bate heartbeat). Acest probe
+citeste `worker_heartbeat` din DB si pica daca ultimul beat e mai vechi decat
+`worker_heartbeat_stale_s` -> Docker restarteaza containerul worker.
+
+Utilizare (compose healthcheck):  python -m app.worker.healthcheck
+Exit 0 = sanatos, 1 = invechit/lipsa.
+"""
+
+from __future__ import annotations
+
+import sys
+from datetime import datetime, timezone
+
+from ..config import Settings, get_settings
+from ..db import get_connection, read_heartbeat
+
+
+def worker_healthy(conn, settings: Settings, *, now: datetime | None = None) -> bool:
+    """True daca ultimul heartbeat e mai proaspat decat pragul de invechire."""
+    hb = read_heartbeat(conn)
+    if hb is None or not hb["last_beat"]:
+        return False
+    try:
+        last = datetime.fromisoformat(hb["last_beat"])
+    except (ValueError, TypeError):
+        return False
+    now = now or datetime.now(timezone.utc)
+    return (now - last).total_seconds() <= settings.worker_heartbeat_stale_s
+
+
+def main() -> int:
+    settings = get_settings()
+    conn = get_connection()
+    try:
+        ok = worker_healthy(conn, settings)
+    finally:
+        conn.close()
+    if not ok:
+        print("[healthcheck] worker invechit sau nepornit", flush=True)
+        return 1
+    return 0
+
+
+if __name__ == "__main__":
+    sys.exit(main())
diff --git a/docker-compose.yml b/docker-compose.yml
index 3b25fe5..4ab5ba5 100644
--- a/docker-compose.yml
+++ b/docker-compose.yml
@@ -1,5 +1,10 @@
 # Gateway RAR AUTOPASS — un container API + un container worker, acelasi image,
 # acelasi volum SQLite persistent (plan.md sect. 4 + 9). restart: always pe ambele.
+#
+# CRITIC: AUTOPASS_CREDS_KEY trebuie PARTAJATA intre api si worker — API cripteaza
+# creds-urile RAR, worker-ul le decripteaza. Chei diferite -> worker nu poate
+# decripta -> submission-uri blocate "creds indisponibile". Seteaz-o in .env
+# (vezi .env.example): compose o citeste automat. Lipsa -> compose pica explicit.
 services:
   api:
     build: .
@@ -11,6 +16,8 @@ services:
     environment:
       AUTOPASS_DB_PATH: /data/autopass.db
       AUTOPASS_RAR_ENV: test
+      AUTOPASS_CREDS_KEY: ${AUTOPASS_CREDS_KEY:?seteaza AUTOPASS_CREDS_KEY in .env (vezi .env.example)}
+      AUTOPASS_REQUIRE_API_KEY: ${AUTOPASS_REQUIRE_API_KEY:-false}
     restart: always
     healthcheck:
       test: ["CMD", "python", "-c", "import urllib.request,sys; sys.exit(0 if urllib.request.urlopen('http://localhost:8000/healthz').status==200 else 1)"]
@@ -26,11 +33,35 @@ services:
     environment:
       AUTOPASS_DB_PATH: /data/autopass.db
       AUTOPASS_RAR_ENV: test
+      AUTOPASS_CREDS_KEY: ${AUTOPASS_CREDS_KEY:?seteaza AUTOPASS_CREDS_KEY in .env (vezi .env.example)}
       # Send dezactivat by default; activeaza pentru proba end-to-end.
       AUTOPASS_WORKER_SEND_ENABLED: "false"
     restart: always
     depends_on:
       - api
+    # T6: probe pe heartbeat-ul din DB — prinde worker-ul AGATAT (proces viu, beat
+    # invechit), pe care restart:always singur nu-l vede. start_period acopera bootul.
+    # ATENTIE: in compose simplu, "unhealthy" doar marcheaza containerul — NU il
+    # restarteaza (restart:always reactioneaza la EXIT). Sidecar-ul `autoheal` de
+    # mai jos vede label-ul si chiar restarteaza worker-ul cand pica probe-ul.
+    labels:
+      autoheal: "true"
+    healthcheck:
+      test: ["CMD", "python", "-m", "app.worker.healthcheck"]
+      interval: 30s
+      timeout: 5s
+      retries: 3
+      start_period: 30s
+
+  # Restarteaza orice container marcat unhealthy cu label autoheal=true (worker-ul
+  # agatat). Alternativa: Docker Swarm (restart on unhealthy nativ).
+  autoheal:
+    image: willfarrell/autoheal:latest
+    restart: always
+    environment:
+      AUTOHEAL_CONTAINER_LABEL: autoheal
+    volumes:
+      - /var/run/docker.sock:/var/run/docker.sock
 
 volumes:
   autopass-data:
diff --git a/docs/plans/plan.md b/docs/plans/plan.md
index 30c531c..e0b5321 100644
--- a/docs/plans/plan.md
+++ b/docs/plans/plan.md
@@ -221,9 +221,15 @@ Nimic din cod nu e scris încă (`app/`, `tools/` nu există). Ordine recomandat
   retry; token expirat → invalidare sesiune + requeue; fără creds (restart) → requeue „indisponibile" (ROAAUTO re-trimite).
   Verify: 10 teste (`tests/test_creds_delivery.py`). **Risc acceptat:** la restart token+creds se pierd → contul re-loghează
   la următorul submission cu creds (degradare per modelul efemer).
-- [ ] **T6 (P2) — worker proces/container propriu supravegheat;** `/healthz` pică → restart. Verify: worker omorât → restart automat.
-- [ ] **T7 (P2) — deploy:** SQLite pe volum persistent numit + backup (singura copie durabilă, re-push scos).
-  Verify: recreare container → coada supraviețuiește.
+- [x] **T6 (P2) — worker supravegheat** ✅ 2026-06-15. `app/worker/healthcheck.py` (probe pe heartbeat-ul din DB: beat
+  mai vechi de `worker_heartbeat_stale_s` → exit 1) cablat în compose ca healthcheck pe serviciul worker. Prinde worker-ul
+  AGĂȚAT (proces viu, beat înghețat), pe care `restart:always` (doar la EXIT) nu-l vede. Sidecar `autoheal` restartează
+  efectiv containerul marcat unhealthy (compose simplu doar marchează, nu restartează). Verify: 3 teste (`tests/test_deploy.py`).
+- [x] **T7 (P2) — deploy** ✅ 2026-06-15. `tools/backup.py` (backup ONLINE via `Connection.backup` — WAL nu se copiază sigur
+  cu `cp`; `--keep N` rotește snapshot-urile) + volum SQLite persistent numit (`autopass-data`, deja în compose). `.env.example`
+  documentează env-urile. **Fix critic descoperit la split-ul în 2 containere:** `AUTOPASS_CREDS_KEY` trebuie PARTAJATĂ
+  api↔worker (altfel worker nu decriptează creds) — acum impusă în compose (`${...:?}` → fail explicit dacă lipsește).
+  Verify: 2 teste (`tests/test_deploy.py`).
 - [ ] **Dashboard** (Jinja2+HTMX) cu stările empty/error/RAR-indisponibil + banner alertă. Apoi `/design-review` pe UI-ul live.
 
 ### De decis ulterior (urmărit, nu blocant)
diff --git a/tests/test_deploy.py b/tests/test_deploy.py
new file mode 100644
index 0000000..858b289
--- /dev/null
+++ b/tests/test_deploy.py
@@ -0,0 +1,117 @@
+"""Teste T6/T7: liveness probe worker + backup online SQLite."""
+
+from __future__ import annotations
+
+import os
+import sqlite3
+import tempfile
+from datetime import datetime, timedelta, timezone
+
+import pytest
+
+
+@pytest.fixture()
+def env(monkeypatch):
+    tmp = tempfile.mkdtemp()
+    monkeypatch.setenv("AUTOPASS_DB_PATH", os.path.join(tmp, "t.db"))
+    from app.config import get_settings
+
+    get_settings.cache_clear()
+    from app.db import get_connection, init_db
+
+    init_db()
+    yield get_connection, get_settings()
+    get_settings.cache_clear()
+
+
+# --------------------------------------------------------------------------- #
+# T6 — worker liveness probe                                                   #
+# --------------------------------------------------------------------------- #
+
+def _set_beat(conn, dt: datetime | None):
+    conn.execute(
+        "UPDATE worker_heartbeat SET last_beat=? WHERE id=1",
+        (dt.isoformat(timespec="seconds") if dt else None,),
+    )
+
+
+def test_healthy_when_beat_fresh(env):
+    from app.worker.healthcheck import worker_healthy
+
+    get_connection, settings = env
+    conn = get_connection()
+    try:
+        _set_beat(conn, datetime.now(timezone.utc))
+        assert worker_healthy(conn, settings) is True
+    finally:
+        conn.close()
+
+
+def test_unhealthy_when_beat_stale(env):
+    from app.worker.healthcheck import worker_healthy
+
+    get_connection, settings = env
+    conn = get_connection()
+    try:
+        stale = datetime.now(timezone.utc) - timedelta(seconds=settings.worker_heartbeat_stale_s + 60)
+        _set_beat(conn, stale)
+        assert worker_healthy(conn, settings) is False
+    finally:
+        conn.close()
+
+
+def test_unhealthy_when_never_started(env):
+    from app.worker.healthcheck import worker_healthy
+
+    get_connection, settings = env
+    conn = get_connection()
+    try:
+        _set_beat(conn, None)  # never started
+        assert worker_healthy(conn, settings) is False
+    finally:
+        conn.close()
+
+
+# --------------------------------------------------------------------------- #
+# T7 — backup online                                                          #
+# --------------------------------------------------------------------------- #
+
+def test_backup_db_roundtrip(env):
+    from tools.backup import backup_db
+
+    get_connection, settings = env
+    conn = get_connection()
+    try:
+        conn.execute(
+            "INSERT INTO submissions (idempotency_key, status, payload_json) VALUES ('bk1','queued','{}')"
+        )
+    finally:
+        conn.close()
+
+    dest = settings.db_path.parent / "backups" / "snap.db"
+    out = backup_db(settings.db_path, dest)
+    assert out.exists()
+    # Copia contine randul scris (backup consistent, nu fisier gol).
+    bk = sqlite3.connect(out)
+    try:
+        n = bk.execute("SELECT COUNT(*) FROM submissions WHERE idempotency_key='bk1'").fetchone()[0]
+    finally:
+        bk.close()
+    assert n == 1
+
+
+def test_backup_prune_keeps_newest(env):
+    from tools.backup import prune
+
+    get_connection, settings = env
+    bdir = settings.db_path.parent / "backups"
+    bdir.mkdir(parents=True, exist_ok=True)
+    # Creeaza 5 snapshot-uri cu nume-timestamp crescator.
+    names = [f"autopass-2026010{i}-000000.db" for i in range(1, 6)]
+    for n in names:
+        (bdir / n).write_bytes(b"x")
+    removed = prune(bdir, keep=2)
+    remaining = sorted(p.name for p in bdir.glob("autopass-*.db"))
+    assert len(remaining) == 2
+    assert remaining == names[-2:]  # cele mai noi doua
+    assert len(removed) == 3
diff --git a/tools/backup.py b/tools/backup.py
new file mode 100644
index 0000000..c7d90ed
--- /dev/null
+++ b/tools/backup.py
@@ -0,0 +1,79 @@
+#!/usr/bin/env python3
+"""Backup online SQLite (T7) — copie consistenta a bazei pe volum persistent.
+
+SQLite in WAL NU se copiaza sigur cu `cp` (WAL-ul poate avea tranzactii necheckpoint-ate).
+Folosim API-ul `Connection.backup` (online, consistent, fara oprirea worker-ului).
+
+Utilizare:
+    python -m tools.backup                      # -> <db_dir>/backups/autopass-YYYYMMDD-HHMMSS.db
+    python -m tools.backup --out /path/snap.db  # destinatie explicita
+    python -m tools.backup --keep 14            # pastreaza ultimele 14, sterge restul
+
+Recomandat: rulat dintr-un cron pe gazda (ex. zilnic), tinta pe volum/montaj separat.
+"""
+
+from __future__ import annotations
+
+import argparse
+import sqlite3
+import sys
+from datetime import datetime, timezone
+from pathlib import Path
+
+from app.config import get_settings
+
+
+def backup_db(src: Path, dest: Path) -> Path:
+    """Backup online (consistent) din `src` in `dest`. Intoarce `dest`."""
+    dest.parent.mkdir(parents=True, exist_ok=True)
+    src_conn = sqlite3.connect(src)
+    try:
+        dst_conn = sqlite3.connect(dest)
+        try:
+            src_conn.backup(dst_conn)
+        finally:
+            dst_conn.close()
+    finally:
+        src_conn.close()
+    return dest
+
+
+def prune(backup_dir: Path, keep: int) -> list[Path]:
+    """Pastreaza cele mai noi `keep` snapshot-uri (dupa nume = timestamp), sterge restul."""
+    snaps = sorted(backup_dir.glob("autopass-*.db"), reverse=True)
+    removed = []
+    for old in snaps[keep:]:
+        old.unlink(missing_ok=True)
+        removed.append(old)
+    return removed
+
+
+def main(argv: list[str] | None = None) -> int:
+    parser = argparse.ArgumentParser(description="Backup online SQLite gateway RAR AUTOPASS")
+    parser.add_argument("--out", type=Path, default=None, help="destinatie explicita (default: <db_dir>/backups/)")
+    parser.add_argument("--keep", type=int, default=0, help="pastreaza ultimele N snapshot-uri (0 = nu sterge)")
+    args = parser.parse_args(argv)
+
+    src = get_settings().db_path
+    if not src.exists():
+        print(f"eroare: baza {src} nu exista", file=sys.stderr)
+        return 1
+
+    if args.out is not None:
+        dest = args.out
+    else:
+        stamp = datetime.now(timezone.utc).strftime("%Y%m%d-%H%M%S")
+        dest = src.parent / "backups" / f"autopass-{stamp}.db"
+
+    backup_db(src, dest)
+    print(f"backup -> {dest} ({dest.stat().st_size} bytes)")
+
+    if args.keep > 0:
+        removed = prune(dest.parent, args.keep)
+        if removed:
+            print(f"sterse {len(removed)} snapshot-uri vechi (pastrez {args.keep})")
+    return 0
+
+
+if __name__ == "__main__":
+    sys.exit(main())