diff --git a/CHANGELOG.md b/CHANGELOG.md
index 0efab46..1d67c1e 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -31,6 +31,8 @@ Follows [Keep a Changelog](https://keepachangelog.com/en/1.0.0/) · versioned wi
   status/cancellation endpoints with timeout and failure states; existing synchronous callers remain compatible.
 - **Optional Background Voice Design.** `/api/voice-design` now accepts `background: true` and exposes bounded job
   status/cancellation endpoints with queue state, timeout and failure handling; existing synchronous callers remain compatible.
+- **Optional Background Casting.** `/api/match-characters-voices` now accepts `background: true` and exposes bounded
+  job status/cancellation endpoints while reusing the existing casting validation, retries, and idempotency behavior.
 - **STT Cancellation Safety.** Queued transcription cancels before consuming the single STT slot, and timed-out
   requests keep the device slot held until the underlying worker actually exits.
 - **Credential Boundary Review.** Provider credentials are now documented as server-only; only the explicitly
diff --git a/TODO.md b/TODO.md
index 8af9f8b..1bfe909 100644
--- a/TODO.md
+++ b/TODO.md
@@ -63,8 +63,8 @@ the first block before investing in architecture or polish.
   a 4 GB VM/container, and measure browser, Docker and temporary-disk usage. Automated browser emulation now proves
   the 4 GB shared-pool cap; physical VM/container and resource measurements remain.
 - [ ] **L — Add background jobs and bounded queues.** Image generation already has the full interactive job path,
-  and transcription plus voice design now have optional ID/status/cancel/timeout paths while preserving synchronous
-  callers. Casting still needs migration to the same user-facing durable job model, plus retries and persistence.
+  and transcription, voice design, and casting now have optional ID/status/cancel/timeout paths while preserving
+  synchronous callers. Durable persistence and retry policy still need to be added.
 - [ ] **L — Run the Book 2 generalization regression.** Confirm speaker-attribution and
   confidence rules are not overfit to *Der Sturm*.
 - [x] **L — Add character forms inside one character record.** `sheet.forms[]` stores bounded alternate appearances,
diff --git a/VERSION b/VERSION
index da9594f..013173a 100644
--- a/VERSION
+++ b/VERSION
@@ -1 +1 @@
-1.22.5
+1.22.6
diff --git a/routes/conversation.py b/routes/conversation.py
index b30440c..1375ed5 100644
--- a/routes/conversation.py
+++ b/routes/conversation.py
@@ -65,6 +65,10 @@ _consistent_image_idempotency: dict[str, str] = {}
 _consistent_image_slots = threading.BoundedSemaphore(2)
 _CAST_IDEMPOTENCY: dict[str, tuple[str, dict, float]] = {}
 _CAST_IDEMPOTENCY_LOCK = threading.Lock()
+_CAST_JOBS: dict[str, dict] = {}
+_CAST_JOB_IDEMPOTENCY: dict[str, tuple[str, float]] = {}
+_CAST_JOB_MAX = 128
+_CAST_JOB_TIMEOUT = 15 * 60
 _CHARACTER_VISUAL_PROMPT_VERSION = "v3"
 _CHARACTER_IDENTITY_FIELDS = (
     "age", "age_estimate", "hair", "eyes", "physical", "build", "skin_tone", "race_species",
@@ -72,6 +76,60 @@ _CHARACTER_IDENTITY_FIELDS = (
 )
 
 
+class _CastJobRequest:
+    """Small internal request adapter used to reuse the validated cast route."""
+
+    def __init__(self, data: dict, headers: dict[str, str]):
+        self._data = data
+        self.headers = headers
+
+    async def json(self):
+        return self._data
+
+
+def _cast_job_set(job_id: str, **fields) -> None:
+    job = _CAST_JOBS.get(job_id)
+    if job is not None:
+        job.update(fields)
+        job["updated"] = time.time()
+
+
+def _cast_job_trim() -> None:
+    if len(_CAST_JOBS) <= _CAST_JOB_MAX:
+        return
+    for job_id, job in sorted(_CAST_JOBS.items(), key=lambda item: item[1].get("updated", 0)):
+        if job.get("done"):
+            _CAST_JOBS.pop(job_id, None)
+            if len(_CAST_JOBS) <= _CAST_JOB_MAX:
+                break
+
+
+def _cast_job_run_sync(data: dict, headers: dict[str, str]):
+    return asyncio.run(match_characters_voices(_CastJobRequest(data, headers)))
+
+
+async def _run_cast_job(job_id: str, data: dict, headers: dict[str, str]) -> None:
+    if _CAST_JOBS.get(job_id, {}).get("cancel_requested"):
+        _cast_job_set(job_id, stage="cancelled", detail="Cancelled", done=True, cancelled=True,
+                      error="Casting cancelled")
+        return
+    _cast_job_set(job_id, stage="running", detail="Matching characters to voices…")
+    call = asyncio.create_task(asyncio.to_thread(_cast_job_run_sync, data, headers))
+    try:
+        result = await asyncio.wait_for(asyncio.shield(call), timeout=_CAST_JOB_TIMEOUT)
+        if _CAST_JOBS.get(job_id, {}).get("cancel_requested"):
+            _cast_job_set(job_id, stage="cancelled", detail="Cancelled", done=True, cancelled=True,
+                          error="Casting cancelled")
+        else:
+            _cast_job_set(job_id, stage="done", detail="Done", done=True, result=result)
+    except asyncio.TimeoutError:
+        call.add_done_callback(lambda task: task.exception() if not task.cancelled() else None)
+        _cast_job_set(job_id, stage="error", detail="Timed out", done=True, error="Voice casting timed out")
+    except Exception as exc:
+        message = str(exc.detail) if isinstance(exc, HTTPException) else str(exc)
+        _cast_job_set(job_id, stage="error", detail="Failed", done=True, error=message)
+
+
 def _character_identity_fingerprint(sheet: dict) -> str:
     sheet = _character_form_sheet(sheet)
     values = {key: sheet.get(key, "") for key in _CHARACTER_IDENTITY_FIELDS}
@@ -618,6 +676,36 @@ async def match_characters_voices(request: Request):
     idempotency_key = str(request.headers.get("idempotency-key") or "").strip()
     if len(idempotency_key) > 200:
         raise HTTPException(400, "Idempotency-Key is too long")
+    if data.get("background") is True:
+        job_fingerprint = hashlib.sha256(json.dumps({k: v for k, v in data.items() if k != "background"}, sort_keys=True, default=str).encode()).hexdigest()
+        if idempotency_key:
+            with _CAST_IDEMPOTENCY_LOCK:
+                previous_job_id, created = _CAST_JOB_IDEMPOTENCY.get(idempotency_key, ("", 0))
+                if previous_job_id and time.time() - created <= 1800:
+                    previous_job = _CAST_JOBS.get(previous_job_id)
+                    if previous_job is not None:
+                        if previous_job.get("request_fingerprint") != job_fingerprint:
+                            raise HTTPException(409, "Idempotency-Key was already used for a different request")
+                        return {"job_id": previous_job_id, "deduplicated": True}
+                _CAST_JOB_IDEMPOTENCY.pop(idempotency_key, None)
+        job_id = uuid.uuid4().hex
+        job_data = dict(data)
+        job_data.pop("background", None)
+        now = time.time()
+        _CAST_JOBS[job_id] = {
+            "job_id": job_id, "stage": "queued", "detail": "Queued", "started": now, "updated": now,
+            "done": False, "cancel_requested": False, "cancelled": False, "error": None, "result": None,
+            "request_fingerprint": job_fingerprint,
+        }
+        if idempotency_key:
+            with _CAST_IDEMPOTENCY_LOCK:
+                _CAST_JOB_IDEMPOTENCY[idempotency_key] = (job_id, now)
+                if len(_CAST_JOB_IDEMPOTENCY) > 256:
+                    oldest = min(_CAST_JOB_IDEMPOTENCY, key=lambda key: _CAST_JOB_IDEMPOTENCY[key][1])
+                    _CAST_JOB_IDEMPOTENCY.pop(oldest, None)
+        _cast_job_trim()
+        asyncio.create_task(_run_cast_job(job_id, job_data, {"idempotency-key": idempotency_key}))
+        return {"job_id": job_id, "stage": "queued"}
     request_fingerprint = hashlib.sha256(json.dumps(data, sort_keys=True, default=str).encode()).hexdigest()
     if idempotency_key:
         now = time.time()
@@ -742,6 +830,30 @@ async def match_characters_voices(request: Request):
     return result
 
 
+@router.get("/api/match-characters-voices/status/{job_id}")
+async def match_characters_voices_status(job_id: str):
+    job = _CAST_JOBS.get(job_id)
+    if job is None:
+        raise HTTPException(404, "Unknown or expired casting job")
+    payload = dict(job)
+    payload["elapsed"] = round(time.time() - payload.pop("started"), 1)
+    payload.pop("cancel_requested", None)
+    payload.pop("request_fingerprint", None)
+    return payload
+
+
+@router.post("/api/match-characters-voices/cancel/{job_id}")
+async def match_characters_voices_cancel(job_id: str):
+    job = _CAST_JOBS.get(job_id)
+    if job is None:
+        raise HTTPException(404, "Unknown or expired casting job")
+    if job.get("done"):
+        return {"ok": True, "already_done": True, "cancelled": bool(job.get("cancelled"))}
+    job["cancel_requested"] = True
+    _cast_job_set(job_id, stage="cancelling", detail="Cancelling…")
+    return {"ok": True, "cancelled": True}
+
+
 def _extract_json_block(text: str) -> str:
     """Pull the first {...} JSON object out of an LLM response."""
     text = re.sub(r"<think>.*?</think>", "", text, flags=re.DOTALL).strip()
diff --git a/tests/test_security.py b/tests/test_security.py
index dbdecb9..0b3f919 100644
--- a/tests/test_security.py
+++ b/tests/test_security.py
@@ -159,6 +159,24 @@ class SecurityValidationTests(unittest.TestCase):
         self.assertIn("same_asset_for_profile_and_reference", result["issues"][0]["issues"])
         self.assertIn("asset_reused_across_characters", result["issues"][0]["issues"])
 
+    def test_cast_background_job_honours_queued_cancellation(self):
+        import routes.conversation as conversation
+        from unittest.mock import patch
+        job_id = "cast-test-job"
+        now = 0.0
+        conversation._CAST_JOBS[job_id] = {
+            "job_id": job_id, "stage": "queued", "detail": "Queued", "started": now, "updated": now,
+            "done": False, "cancel_requested": True, "cancelled": False,
+        }
+        try:
+            with patch.object(conversation, "match_characters_voices") as match:
+                asyncio.run(conversation._run_cast_job(job_id, {}, {}))
+            self.assertTrue(conversation._CAST_JOBS[job_id]["done"])
+            self.assertEqual(conversation._CAST_JOBS[job_id]["stage"], "cancelled")
+            match.assert_not_called()
+        finally:
+            conversation._CAST_JOBS.pop(job_id, None)
+
     def test_fish_cache_eviction_is_bounded(self):
         from pathlib import Path
         from tempfile import TemporaryDirectory