fix: clarify ForceEndTurn deployment behavior

.agents/skills/deepgram-python-conversational-stt/SKILL.md

@@ -91,7 +91,7 @@ with client.listen.v2.connect(
 
 **No `language` parameter** on v2 — language is implied by model (`flux-general-en`) or hinted via `language_hint` on multi.
 
-For application-controlled turns, use `eot_threshold="1.0"` with a sufficiently large `eot_timeout_ms`, then call `conn.send_force_end_turn()` for the active turn. ForceEndTurn requires deployment enablement; see `examples/16-transcription-force-end-turn.py`.
+For application-controlled turns, use `eot_threshold="1.0"` with a sufficiently large `eot_timeout_ms`, then call `conn.send_force_end_turn()` for the active turn. ForceEndTurn is available on Deepgram-hosted deployments, including EU and AU. On self-hosted deployments, it requires the 2026-09-15 release or later with `listen_v2 = true` and `listen_v2_force_end_turn = true` under `[features]`; see `examples/16-transcription-force-end-turn.py`.
 
 ## Events (server → client)
 

.fernignore

@@ -65,7 +65,7 @@ src/deepgram/speak/v2/socket_client.py
 
 # Agent TTS provider expressivity must reject fractional values rather than
 # allowing Pydantic v1 to truncate them before the API sees the request.
-# [temporarily frozen — remove when Fern emits a strict integer]
+# [temporarily frozen — remove when Pydantic v1 support is dropped]
 src/deepgram/types/speak_settings_v1provider.py
 src/deepgram/types/deepgram.py
 
@@ -268,6 +268,7 @@ tests/custom/test_flux_tts_controls.py
 tests/custom/test_topics_intents_v7_compat.py
 tests/custom/test_transport.py
 tests/custom/test_websocket_streaming_coverage.py
+tests/custom/test_voice_agent_force_end_turn_example.py
 tests/typecheck/compat_aliases.py
 
 # Wire test with restored compatibility coverage for legacy create-key request alias

AGENTS.md

@@ -43,6 +43,7 @@ Current permanently frozen files:
 - `tests/custom/test_regen_response_fields.py` — hand-written regression test guarding the response fields added by the 2026-08-18 regen: `ListenV1Response/ResultsMetadata.diarize_info` (`model_uuid`, `arch`), per-word `speaker` / `speaker_confidence`, and `ListenV2TurnInfo.trigger`. Same rationale as `test_latency_report_stt_compat.py`: `UncheckedBaseModel` + `skip_validation` makes response parsing lenient, so a future spec removal would silently turn these into `None` rather than failing. Fixtures are real captured payloads; the `trigger` cases pin the parsing contract for the values the server sends (`model` / `manual` / `timeout`) plus the absent case, which is what a deployment without the server-gated `ForceEndTurn` feature returns. Live coverage of the same field lives in `tests/manual/listen/v2/force_end_turn/`
 - `tests/custom/test_secure_logging.py` — hand-written regression test that the `websockets` Authorization-header DEBUG logs are redacted (API key never logged in clear text)
 - `tests/custom/test_speak_v2_interrupt_configure.py` — hand-written coverage for the Speak V2 barge-in / mid-stream reconfigure surface (`send_interrupt`, `send_configure`, `SpeechInterrupted`, `ConfigureSuccess`/`ConfigureFailure`, and the `speed`/`expressivity` connect params)
+- `tests/custom/test_voice_agent_force_end_turn_example.py` — hand-written regression coverage that the Voice Agent ForceEndTurn example waits for a rejection queued after `AgentAudioDone`
 - `tests/custom/test_text_builder.py`, `tests/custom/test_transport.py` — hand-written tests
 - `tests/custom/test_flux_tts_controls.py`, `tests/custom/test_topics_intents_v7_compat.py` — hand-written regression coverage for Flux TTS controls and direct Topics/Intents response compatibility
 - `tests/typecheck/compat_aliases.py` — hand-written mypy `assert_type` coverage for backward-compatible alias TypedDicts
@@ -64,10 +65,10 @@ Current temporarily frozen files:
 - `.gitignore` — Fern generates a baseline version and has regenerated it before (`fdcce88`, `ad93815`, `4bee463`). We hand-add the local coverage artifact ignores (`.coverage`, `htmlcov/`, `coverage.xml`). Before each regen, unfreeze and re-diff so Fern's own additions are picked up, then re-apply the three coverage lines.
 - `src/deepgram/speak/v1/socket_client.py` — optional message param defaults, broad exception catch
 - `src/deepgram/speak/v2/socket_client.py` — same (optional `send_flush`/`send_close`/`send_interrupt` defaults, broad exception catch); new websocket TTS client added in the 2026-07-08 regen. `send_interrupt` carries no required payload so it takes the same optional-default treatment as the other control sends; `send_configure` deliberately keeps its required argument (a Configure with no settings is meaningless)
-- `src/deepgram/types/speak_settings_v1provider.py`, `src/deepgram/types/deepgram.py` — validate Agent TTS `expressivity` as `pydantic.StrictInt` so Pydantic v1 rejects fractional values instead of truncating them before they reach the API. Regression coverage in `tests/custom/test_socket_client_shims.py`. Unfreeze when Fern emits a strict integer.
 - `src/deepgram/listen/v1/socket_client.py` — same
 - `src/deepgram/listen/v2/socket_client.py` — same (broad except, optional `send_close_stream` default). As of the 2026-08-11 regen the generator properly types `send_configure(ListenV2Configure)` and puts `ListenV2ConfigureSuccess` in the response Union, so those are taken from the generator; the only `send_configure` patch retained is runtime tolerance for a raw dict (sent verbatim) for back-compat with pre-typed-model callers
 - `src/deepgram/agent/v1/socket_client.py` — same + `_sanitize_numeric_types`
+- `src/deepgram/types/speak_settings_v1provider.py`, `src/deepgram/types/deepgram.py` — validate Agent TTS `expressivity` as `pydantic.StrictInt` so Pydantic v1 rejects fractional values instead of truncating them before they reach the API. Regression coverage in `tests/custom/test_socket_client_shims.py`. Unfreeze when Pydantic v1 support is dropped.
 - `src/deepgram/agent/v1/types/agent_v1settings_agent_context.py`, `src/deepgram/agent/v1/types/agent_v1settings_agent.py`, `src/deepgram/agent/v1/types/agent_v1settings.py`, `src/deepgram/agent/v1/requests/agent_v1settings_agent_context.py`, `src/deepgram/agent/v1/requests/agent_v1settings_agent.py`, `src/deepgram/agent/v1/requests/agent_v1settings.py` — backward-compat patches for the 2026-05-05 Agent Settings schema restructure. These preserve callable `AgentV1SettingsAgent(...)`, keep `AgentV1Settings.agent` accepting both that wrapper and `agent_id` strings, restore the legacy request TypedDict shapes, remap legacy `messages=[...]` / nested `context=AgentV1SettingsAgentContext(messages=[...])` usage into the new `context={"messages": [...]}` wire shape, and keep read-side `obj.messages` access working.
 - `src/deepgram/core/api_error.py`, `src/deepgram/core/parse_error.py` — credential redaction. Every websocket `connect()` path raises `ApiError(headers=dict(headers), ...)` with the full request headers, and both error types stringify that dict, so an unredacted `Authorization` reached `str(e)`, tracebacks, log aggregators and error trackers (which serialise attributes as well as the message). Both now mask credential values at construction via `_secure_logging.redact_sensitive_headers`, preserving non-sensitive headers (`dg-request-id`) for debugging. This is the same threat `_secure_logging.py` covers for the `websockets` DEBUG handshake logs, via the other path to it. Regression coverage in `tests/custom/test_api_error_redaction.py`. Unfreeze if the generator starts redacting credentials itself.
 - `src/deepgram/core/query_encoder.py` — coerces Python bools to lowercase `"true"`/`"false"` before they reach `urllib.parse.urlencode` (which would otherwise produce `"True"`/`"False"` via `str()` and break websocket query strings). Only the four `*/connect()` paths call `urlencode`; HTTP raw clients hand params to httpx, which lowercases bools itself, so the patch is a no-op for the HTTP path. Once Fern's websocket codegen normalizes bools (or the spec types these as `boolean` end-to-end), this can be unfrozen.

examples/16-transcription-force-end-turn.py

@@ -21,9 +21,15 @@
 push-to-talk button being released. The connection stays open after a forced end: the
 turn index advances and transcription continues.
 
-Note: ForceEndTurn requires server-side enablement and is not available on every
-deployment. Where it is not enabled the server replies UNPARSABLE_CLIENT_MESSAGE and
-closes the connection; this example reports that and exits.
+ForceEndTurn is available on Deepgram-hosted deployments, including EU and AU. On
+self-hosted deployments, it requires the 2026-09-15 release or later with
+the following [features] settings:
+
+    listen_v2 = true
+    listen_v2_force_end_turn = true
+
+Unsupported deployments reply UNPARSABLE_CLIENT_MESSAGE and close the connection; this
+example reports that and exits.
 """
 
 import os
@@ -148,7 +154,8 @@ def send_audio() -> None:
     # before it could print anything.
     if feature_disabled.is_set():
         print("\nForceEndTurn is not enabled on this deployment.")
-        print("Set DEEPGRAM_BASE_URL to a deployment that has the feature.")
+        print("Self-hosted deployments need the 2026-09-15 release or later with")
+        print("listen_v2 = true and listen_v2_force_end_turn = true under [features].")
 
 except Exception as e:
     print(f"Error: {type(e).__name__}: {e}")

examples/32-voice-agent-force-end-turn.py

@@ -31,11 +31,36 @@
 load_dotenv()
 
 AUDIO_PATH = Path(__file__).parent / "fixtures" / "audio.wav"
+FORCE_END_TURN_REJECTION_GRACE_SECONDS = 0.5
+
+
+def wait_for_force_end_turn_outcome(
+    agent_finished: threading.Event,
+    force_end_turn_rejected: threading.Event,
+    *,
+    timeout_seconds: float,
+) -> bool:
+    """Return whether ForceEndTurn completed without a queued rejection."""
+    deadline = time.monotonic() + timeout_seconds
+    while True:
+        if force_end_turn_rejected.is_set():
+            return False
+
+        remaining = deadline - time.monotonic()
+        if remaining <= 0:
+            raise TimeoutError("Timed out waiting for the agent response")
+        if agent_finished.wait(min(0.1, remaining)):
+            # AgentAudioDone can precede FORCE_END_TURN_UNSUPPORTED in the socket reader.
+            return not force_end_turn_rejected.wait(
+                min(FORCE_END_TURN_REJECTION_GRACE_SECONDS, remaining)
+            )
 
 
 def main() -> None:
     user_started = threading.Event()
     agent_finished = threading.Event()
+    force_end_turn_rejected = threading.Event()
+    force_end_turn_error: str | None = None
 
     settings = AgentV1Settings(
         audio=AgentV1SettingsAudio(input=AgentV1SettingsAudioInput(encoding="linear16", sample_rate=44100)),
@@ -56,6 +81,7 @@ def main() -> None:
     with DeepgramClient().agent.v1.connect() as agent:
 
         def on_message(message: object) -> None:
+            nonlocal force_end_turn_error
             message_type = getattr(message, "type", None)
             if message_type == "UserStartedSpeaking":
                 user_started.set()
@@ -67,6 +93,9 @@ def on_message(message: object) -> None:
                 print("AgentAudioDone received")
             elif message_type in {"Warning", "Error"}:
                 print(f"{message_type}: {message.code} - {message.description}")
+                if message.code == "FORCE_END_TURN_UNSUPPORTED":
+                    force_end_turn_error = f"{message.code}: {message.description}"
+                    force_end_turn_rejected.set()
 
         agent.on(EventType.MESSAGE, on_message)
         agent.on(EventType.ERROR, lambda error: print(f"Connection error: {error}"))
@@ -88,8 +117,12 @@ def on_message(message: object) -> None:
 
         print("Sending ForceEndTurn")
         agent.send_force_end_turn()
-        if not agent_finished.wait(15):
-            raise TimeoutError("Timed out waiting for the agent response")
+        if not wait_for_force_end_turn_outcome(
+            agent_finished,
+            force_end_turn_rejected,
+            timeout_seconds=15,
+        ):
+            raise RuntimeError(f"ForceEndTurn failed: {force_end_turn_error}")
 
 
 if __name__ == "__main__":

tests/custom/test_voice_agent_force_end_turn_example.py

@@ -0,0 +1,45 @@
+import importlib.util
+import sys
+import threading
+import types
+from pathlib import Path
+
+EXAMPLE_PATH = Path(__file__).parents[2] / "examples" / "32-voice-agent-force-end-turn.py"
+
+
+class RejectionDuringGraceWait(threading.Event):
+    def __init__(self):
+        super().__init__()
+        self.wait_calls = 0
+
+    def wait(self, _timeout):
+        self.wait_calls += 1
+        self.set()
+        return True
+
+
+def load_example(monkeypatch):
+    dotenv = types.ModuleType("dotenv")
+    dotenv.load_dotenv = lambda: None
+    monkeypatch.setitem(sys.modules, "dotenv", dotenv)
+
+    spec = importlib.util.spec_from_file_location("voice_agent_force_end_turn_example", EXAMPLE_PATH)
+    assert spec is not None
+    assert spec.loader is not None
+    module = importlib.util.module_from_spec(spec)
+    spec.loader.exec_module(module)
+    return module
+
+
+def test_agent_audio_done_waits_for_late_force_end_turn_rejection(monkeypatch):
+    example = load_example(monkeypatch)
+    agent_finished = threading.Event()
+    force_end_turn_rejected = RejectionDuringGraceWait()
+    agent_finished.set()
+
+    assert not example.wait_for_force_end_turn_outcome(
+        agent_finished,
+        force_end_turn_rejected,
+        timeout_seconds=1,
+    )
+    assert force_end_turn_rejected.wait_calls == 1

tests/manual/listen/v2/force_end_turn/async.py

@@ -5,8 +5,10 @@
 step 1 of `main.py`: force an in-progress turn to end and confirm the resulting EndOfTurn
 reports trigger="manual". See `main.py` for the full end-of-turn control matrix.
 
-ForceEndTurn is gated per deployment; where it is not enabled this reports SKIP rather
-than failing. Point it at a deployment that has the feature with DEEPGRAM_BASE_URL.
+ForceEndTurn is available on Deepgram-hosted deployments, including EU and AU. On
+self-hosted deployments, it requires the 2026-09-15 release or later with
+`listen_v2 = true` and `listen_v2_force_end_turn = true` under [features]. Unsupported
+deployments report SKIP.
 
 Requires DEEPGRAM_API_KEY. Run with:
 
@@ -114,7 +116,8 @@ def on_message(message: Any) -> None:
     if gated:
         print("  SKIP: ForceEndTurn is not enabled on this deployment")
         print(f"        ({errors[0]})")
-        print("        Set DEEPGRAM_BASE_URL to a deployment that has the feature.")
+        print("        Self-hosted deployments need the 2026-09-15 release or later with")
+        print("        listen_v2 = true and listen_v2_force_end_turn = true under [features].")
         return
     if errors:
         print(f"  FAIL: {errors[0]}")

tests/manual/listen/v2/force_end_turn/main.py

@@ -21,10 +21,15 @@
        ForceEndTurn becomes the only way to close a turn. This is the combination to
        reach for when the application owns turn boundaries.
 
-ForceEndTurn is gated per deployment. Where it is not enabled the server replies
-UNPARSABLE_CLIENT_MESSAGE ("not enabled on this deployment") and closes the connection;
-this script reports SKIP rather than failing. Point it at a deployment that has the
-feature with DEEPGRAM_BASE_URL.
+ForceEndTurn is available on Deepgram-hosted deployments, including EU and AU. On
+self-hosted deployments, it requires the 2026-09-15 release or later with
+the following [features] settings:
+
+    listen_v2 = true
+    listen_v2_force_end_turn = true
+
+Unsupported deployments reply UNPARSABLE_CLIENT_MESSAGE ("not enabled on this deployment")
+and close the connection; this script reports SKIP rather than failing.
 
 Requires DEEPGRAM_API_KEY. Run with:
 
@@ -173,7 +178,8 @@ def main() -> None:
     if gated:
         print("  SKIP: ForceEndTurn is not enabled on this deployment")
         print(f"        ({errors[0]})")
-        print("        Set DEEPGRAM_BASE_URL to a deployment that has the feature.")
+        print("        Self-hosted deployments need the 2026-09-15 release or later with")
+        print("        listen_v2 = true and listen_v2_force_end_turn = true under [features].")
         return
     if errors:
         raise AssertionError(errors[0])