Speakr is a personal, self-hosted web application designed for transcribing audio recordings
Python
Provider-specific request parameters for the text and chat models (TEXT_MODEL_EXTRA_BODY, CHAT_MODEL_EXTRA_BODY)
Murtaza Nasir187260bf0d13
diff --git a/config/env.asr.example b/config/env.asr.exampleindex 25cddca4..8a283801 100644--- a/config/env.asr.example+++ b/config/env.asr.example@@ -68,6 +68,17 @@ GPT5_VERBOSITY=medium# CHAT_GPT5_REASONING_EFFORT=medium# CHAT_GPT5_VERBOSITY=medium+# --- Provider-Specific Request Parameters (Optional) ---+# A JSON object added to every request to the model, for settings such as+# reasoning effort that differ between providers. Examples:+# OpenRouter: {"reasoning": {"effort": "low"}}+# OpenAI: {"reasoning_effort": "minimal"}+# vLLM or llama.cpp, Qwen3: {"chat_template_kwargs": {"enable_thinking": false}}+# CHAT_MODEL_EXTRA_BODY is used only for a separately configured chat model.+# See docs/admin-guide/model-configuration.md.+# TEXT_MODEL_EXTRA_BODY={"reasoning": {"effort": "low"}}+# CHAT_MODEL_EXTRA_BODY={"reasoning": {"effort": "low"}}+# --- LLM Timeout & Retry ---# Read timeout for LLM requests in seconds (default 600). Increase for local# models (Ollama, vLLM) that need more time for long transcripts.diff --git a/config/env.transcription.example b/config/env.transcription.exampleindex 1c8e635b..031d167a 100644--- a/config/env.transcription.example+++ b/config/env.transcription.example@@ -77,6 +77,17 @@ GPT5_VERBOSITY=medium# CHAT_MODEL_BASE_URL=https://openrouter.ai/api/v1# CHAT_MODEL_NAME=openai/gpt-4o+# --- Provider-Specific Request Parameters (Optional) ---+# A JSON object added to every request to the model, for settings such as+# reasoning effort that differ between providers. Examples:+# OpenRouter: {"reasoning": {"effort": "low"}}+# OpenAI: {"reasoning_effort": "minimal"}+# vLLM or llama.cpp, Qwen3: {"chat_template_kwargs": {"enable_thinking": false}}+# CHAT_MODEL_EXTRA_BODY is used only for a separately configured chat model.+# See docs/admin-guide/model-configuration.md.+# TEXT_MODEL_EXTRA_BODY={"reasoning": {"effort": "low"}}+# CHAT_MODEL_EXTRA_BODY={"reasoning": {"effort": "low"}}+# =============================================================================# CONNECTOR SELECTION (Auto-detected if not set)# =============================================================================diff --git a/config/env.whisper.example b/config/env.whisper.exampleindex 8179569d..29537c5b 100644--- a/config/env.whisper.example+++ b/config/env.whisper.example@@ -70,6 +70,17 @@ GPT5_VERBOSITY=medium# CHAT_GPT5_REASONING_EFFORT=medium# CHAT_GPT5_VERBOSITY=medium+# --- Provider-Specific Request Parameters (Optional) ---+# A JSON object added to every request to the model, for settings such as+# reasoning effort that differ between providers. Examples:+# OpenRouter: {"reasoning": {"effort": "low"}}+# OpenAI: {"reasoning_effort": "minimal"}+# vLLM or llama.cpp, Qwen3: {"chat_template_kwargs": {"enable_thinking": false}}+# CHAT_MODEL_EXTRA_BODY is used only for a separately configured chat model.+# See docs/admin-guide/model-configuration.md.+# TEXT_MODEL_EXTRA_BODY={"reasoning": {"effort": "low"}}+# CHAT_MODEL_EXTRA_BODY={"reasoning": {"effort": "low"}}+# --- LLM Timeout & Retry ---# Read timeout for LLM requests in seconds (default 600). Increase for local# models (Ollama, vLLM) that need more time for long transcripts.diff --git a/config/env.whisperx.example b/config/env.whisperx.exampleindex e43978bf..3e72a9ea 100644--- a/config/env.whisperx.example+++ b/config/env.whisperx.example@@ -84,6 +84,17 @@ GPT5_VERBOSITY=medium# CHAT_GPT5_REASONING_EFFORT=medium# CHAT_GPT5_VERBOSITY=medium+# --- Provider-Specific Request Parameters (Optional) ---+# A JSON object added to every request to the model, for settings such as+# reasoning effort that differ between providers. Examples:+# OpenRouter: {"reasoning": {"effort": "low"}}+# OpenAI: {"reasoning_effort": "minimal"}+# vLLM or llama.cpp, Qwen3: {"chat_template_kwargs": {"enable_thinking": false}}+# CHAT_MODEL_EXTRA_BODY is used only for a separately configured chat model.+# See docs/admin-guide/model-configuration.md.+# TEXT_MODEL_EXTRA_BODY={"reasoning": {"effort": "low"}}+# CHAT_MODEL_EXTRA_BODY={"reasoning": {"effort": "low"}}+# --- Transcription Service (WhisperX ASR Endpoint) ---# New connector architecture auto-detects ASR mode when ASR_BASE_URL is set.# USE_ASR_ENDPOINT=true is deprecated but still works for backwards compatibility.diff --git a/docs/admin-guide/model-configuration.md b/docs/admin-guide/model-configuration.mdindex 32b38e2c..c93745ec 100644--- a/docs/admin-guide/model-configuration.md+++ b/docs/admin-guide/model-configuration.md@@ -376,6 +376,33 @@ After changing model configuration:4. Test the chat feature5. Monitor logs for any errors or warnings+## Provider-Specific Request Parameters++Reasoning ("thinking") is controlled differently by each provider, and some models think by default, which makes summaries and titles slower and more expensive. `TEXT_MODEL_EXTRA_BODY` holds a JSON object whose fields are added to every request to the text model. `CHAT_MODEL_EXTRA_BODY` is the same setting for a separately configured chat model.++```bash+# Turn off thinking for a Qwen3 model on vLLM, SGLang or llama.cpp+TEXT_MODEL_EXTRA_BODY={"chat_template_kwargs": {"enable_thinking": false}}+```++The fields are sent unchanged, so they must be ones your provider and model accept:++| Provider | Example |+|----------|---------|+| OpenRouter, any reasoning model | `{"reasoning": {"effort": "low"}}`, or `{"reasoning": {"enabled": false}}` to turn reasoning off |+| OpenAI (o-series, GPT-5) | `{"reasoning_effort": "minimal"}` |+| vLLM, SGLang or llama.cpp, Qwen3 and similar models | `{"chat_template_kwargs": {"enable_thinking": false}}` |+| vLLM, gpt-oss | `{"reasoning_effort": "low"}` |++The name of the `chat_template_kwargs` field depends on the model's chat template and is given on the model card.++- The fields are added after Speakr's own parameters and replace any with the same name, including the GPT-5 settings and `max_tokens`.+- When chat uses the text model, `TEXT_MODEL_EXTRA_BODY` applies to chat as well. A separately configured chat model does not inherit it.+- In `.env`, the JSON is written without surrounding quotes.+- If the value is not a JSON object, requests to that model fail with an error that names the variable.+- When a provider rejects a request, the log lists the fields added by this setting. Some servers ignore fields they do not recognise, so compare response times or token counts before and after a change to confirm its effect.+- Reasoning tokens are counted as completion tokens in the token statistics. The reasoning text is not stored with summaries, titles or chat answers.+## Environment Variables Reference```bash@@ -417,6 +444,11 @@ CHAT_MODEL_NAME=openai/gpt-4o# Chat-specific GPT-5 settings (optional - falls back to GPT5_* if not set)CHAT_GPT5_REASONING_EFFORT=medium # minimal, low, medium, highCHAT_GPT5_VERBOSITY=medium # low, medium, high++# Provider-specific request parameters, a JSON object (optional)+# See Provider-Specific Request Parameters above+TEXT_MODEL_EXTRA_BODY={"reasoning": {"effort": "low"}}+CHAT_MODEL_EXTRA_BODY={"reasoning": {"effort": "low"}}```## Prefix-Cache-Friendly Prompts (`PREFIX_CACHE_OPTIMIZED_PROMPTS`)diff --git a/src/api/recordings.py b/src/api/recordings.pyindex 890ac93e..7b75dddf 100644--- a/src/api/recordings.py+++ b/src/api/recordings.py@@ -3195,8 +3195,6 @@ def chat_incognito():from src.services.recording_chat import build_chat_messagesmessages, _ = build_chat_messages(recording, current_user, user_message, message_history)- # Get model info- chat_model = os.environ.get('TEXT_MODEL_NAME', os.environ.get('OPENAI_CHAT_MODEL', 'gpt-4o-mini'))user_id = current_user.idcurrent_app.logger.info(f"[Incognito Chat] User {user_id} sending message")@@ -3207,12 +3205,8 @@ def chat_incognito():def generate():"""Stream the chat response."""try:- response = chat_client.chat.completions.create(- model=chat_model,- messages=messages,- temperature=chat_temperature,- stream=True- )+ # No user_id, so nothing is tracked for incognito+ response = call_chat_completion(messages, temperature=chat_temperature, stream=True)for chunk in response:if chunk.choices and len(chunk.choices) > 0:diff --git a/src/services/llm.py b/src/services/llm.pyindex 1ac19315..678da7bb 100644--- a/src/services/llm.py+++ b/src/services/llm.py@@ -8,7 +8,7 @@ import jsonimport timeimport loggingimport httpx-from openai import OpenAI, APITimeoutError+from openai import OpenAI, APITimeoutError, BadRequestError# Use standard logging instead of current_app.logger for context independencelogger = logging.getLogger(__name__)@@ -84,6 +84,30 @@ CHAT_MODEL_NAME = os.environ.get("CHAT_MODEL_NAME")CHAT_GPT5_REASONING_EFFORT = os.environ.get("CHAT_GPT5_REASONING_EFFORT")CHAT_GPT5_VERBOSITY = os.environ.get("CHAT_GPT5_VERBOSITY")+# Provider-specific request parameters, as a JSON object merged into the body+# of every request to the model (OpenAI SDK extra_body). Reasoning controls+# differ by provider and model (reasoning_effort, OpenRouter's reasoning+# object, chat_template_kwargs on vLLM), so they are set here, not mapped.+def _parse_extra_body(env_var):+ """Return (parameters, error) for a JSON object in env_var."""+ raw = os.environ.get(env_var, "").strip()+ if not raw:+ return {}, None+ try:+ value = json.loads(raw)+ except ValueError as e:+ return {}, f"{env_var} is not valid JSON ({e})"+ if not isinstance(value, dict):+ return {}, f"{env_var} must be a JSON object, such as {{\"reasoning_effort\": \"low\"}}"+ return value, None+++TEXT_MODEL_EXTRA_BODY, TEXT_MODEL_EXTRA_BODY_ERROR = _parse_extra_body("TEXT_MODEL_EXTRA_BODY")+CHAT_MODEL_EXTRA_BODY, CHAT_MODEL_EXTRA_BODY_ERROR = _parse_extra_body("CHAT_MODEL_EXTRA_BODY")+for _error in (TEXT_MODEL_EXTRA_BODY_ERROR, CHAT_MODEL_EXTRA_BODY_ERROR):+ if _error:+ logger.error(f"{_error}; requests to that model will fail until it is corrected.")+# Streaming options - disable for LLM servers that don't support OpenAI's stream_optionsENABLE_STREAM_OPTIONS = os.environ.get("ENABLE_STREAM_OPTIONS", "true").lower() == "true"@@ -130,6 +154,11 @@ def get_chat_config():# A separate chat model does not inherit GPT5_FORCE_MODE: it may# be a different (non-GPT-5) model than the main text model.'gpt5_force': _gpt5_mode_forced("CHAT_GPT5_FORCE_MODE"),+ # Like GPT5_FORCE_MODE, the text model's parameters are not+ # inherited: they may not suit a different model or server.+ 'extra_body': CHAT_MODEL_EXTRA_BODY,+ 'extra_body_variable': 'CHAT_MODEL_EXTRA_BODY',+ 'extra_body_error': CHAT_MODEL_EXTRA_BODY_ERROR,}return {'api_key': TEXT_MODEL_API_KEY,@@ -138,6 +167,9 @@ def get_chat_config():'gpt5_reasoning_effort': os.environ.get("GPT5_REASONING_EFFORT", "medium"),'gpt5_verbosity': os.environ.get("GPT5_VERBOSITY", "medium"),'gpt5_force': _gpt5_mode_forced(),+ 'extra_body': TEXT_MODEL_EXTRA_BODY,+ 'extra_body_variable': 'TEXT_MODEL_EXTRA_BODY',+ 'extra_body_error': TEXT_MODEL_EXTRA_BODY_ERROR,}@@ -208,6 +240,26 @@ except Exception as chat_client_init_e:chat_client = client+def _add_extra_body(completion_args, extra_body, error):+ """Merge the configured request parameters into completion_args.++ They are applied last, so they take precedence over the parameters set+ above (temperature, max_tokens, the GPT-5 settings)."""+ if error:+ raise ValueError(error)+ if extra_body:+ completion_args["extra_body"] = dict(extra_body)+++def _log_rejected_parameters(error, variable, extra_body):+ """Name the request parameters setting when the provider rejects a request."""+ if extra_body and isinstance(error, BadRequestError):+ logger.error(+ f"The model server rejected the request. {variable} adds "+ f"{', '.join(sorted(extra_body))}; check that the server accepts these parameters."+ )++def is_gpt5_model(model_name):"""Check if the model is a GPT-5 series model that requires special API parameters.@@ -313,6 +365,8 @@ def call_llm_completion(messages, temperature=0.7, response_format=None, stream=if response_format:completion_args["response_format"] = response_format+ _add_extra_body(completion_args, TEXT_MODEL_EXTRA_BODY, TEXT_MODEL_EXTRA_BODY_ERROR)+# Make the resolved max_tokens visible in logs so users can confirm# their SUMMARY_MAX_TOKENS / CHAT_MAX_TOKENS / etc. settings actually# took effect for a given operation. The "key" used here is@@ -373,6 +427,7 @@ def call_llm_completion(messages, temperature=0.7, response_format=None, stream=raiseexcept Exception as e:logger.error(f"LLM API call failed: {e}")+ _log_rejected_parameters(e, 'TEXT_MODEL_EXTRA_BODY', TEXT_MODEL_EXTRA_BODY)raise@@ -403,9 +458,6 @@ def call_chat_completion(messages, temperature=0.7, response_format=None, streamif not effective_client:raise ValueError("Chat LLM client not initialized")- if not chat_config['api_key']:- raise ValueError("Chat model API key not configured")-# Check budget before making the callif user_id and operation_type:try:@@ -461,6 +513,8 @@ def call_chat_completion(messages, temperature=0.7, response_format=None, streamif tool_choice:completion_args["tool_choice"] = tool_choice+ _add_extra_body(completion_args, chat_config['extra_body'], chat_config['extra_body_error'])+# Visibility: surface the resolved budget per call so admins can# confirm CHAT_MAX_TOKENS or per-call overrides took effect.budget_key = 'max_completion_tokens' if using_gpt5 else 'max_tokens'@@ -514,6 +568,7 @@ def call_chat_completion(messages, temperature=0.7, response_format=None, streamraiseexcept Exception as e:logger.error(f"Chat LLM API call failed: {e}")+ _log_rejected_parameters(e, chat_config['extra_body_variable'], chat_config['extra_body'])raisediff --git a/src/tasks/processing.py b/src/tasks/processing.pyindex 98db89cd..fb68f0f6 100644--- a/src/tasks/processing.py+++ b/src/tasks/processing.py@@ -2745,9 +2745,8 @@ Title:"""if user_output_language:system_message_content += f" Ensure your response is in {user_output_language}."- # Use call_llm_completion without user_id tracking for incognito- completion = client.chat.completions.create(- model=TEXT_MODEL_NAME,+ # No user_id, so nothing is tracked for incognito+ completion = call_llm_completion(messages=[{"role": "system", "content": system_message_content},{"role": "user", "content": prompt_text}@@ -2820,9 +2819,8 @@ Summarization Instructions:current_app.logger.info(f"[Incognito] Generating summary...")- # Use client directly without user tracking for incognito- completion = client.chat.completions.create(- model=TEXT_MODEL_NAME,+ # No user_id, so nothing is tracked for incognito+ completion = call_llm_completion(messages=[{"role": "system", "content": system_message_content},{"role": "user", "content": prompt_text}diff --git a/tests/test_cov_llm.py b/tests/test_cov_llm.pyindex 2535778a..c492476b 100644--- a/tests/test_cov_llm.py+++ b/tests/test_cov_llm.py@@ -463,15 +463,16 @@ def test_call_chat_completion_no_client(monkeypatch):call_chat_completion([{"role": "user", "content": "hi"}])-def test_call_chat_completion_no_api_key(monkeypatch):+def test_call_chat_completion_without_api_key(monkeypatch):+ """A local server often needs no key (#403), for chat as for summaries."""fake = FakeClient(return_value=make_completion("ok"))monkeypatch.setattr(llm, "chat_client", fake)monkeypatch.setattr(llm, "client", fake)monkeypatch.setattr(llm, "CHAT_MODEL_API_KEY", None)monkeypatch.setattr(llm, "CHAT_MODEL_NAME", None)monkeypatch.setattr(llm, "TEXT_MODEL_API_KEY", None)- with pytest.raises(ValueError, match="API key not configured"):- call_chat_completion([{"role": "user", "content": "hi"}])+ response = call_chat_completion([{"role": "user", "content": "hi"}])+ assert response.choices[0].message.content == "ok"def test_call_chat_completion_success(monkeypatch):diff --git a/tests/test_llm_extra_body.py b/tests/test_llm_extra_body.pynew file mode 100644index 00000000..52af9c2d--- /dev/null+++ b/tests/test_llm_extra_body.py@@ -0,0 +1,138 @@+"""Provider-specific request parameters (TEXT_MODEL_EXTRA_BODY and+CHAT_MODEL_EXTRA_BODY), merged into every request as extra_body."""++from types import SimpleNamespace+from unittest.mock import MagicMock++import httpx+import openai+import pytest++import src.services.llm as llm+++def _completion(content="ok"):+ return SimpleNamespace(choices=[SimpleNamespace(message=SimpleNamespace(content=content),+ finish_reason="stop")], usage=None)+++def _client(monkeypatch, *names, **kwargs):+ create = MagicMock(return_value=_completion(), **kwargs)+ fake = SimpleNamespace(chat=SimpleNamespace(completions=SimpleNamespace(create=create)))+ for name in names:+ monkeypatch.setattr(llm, name, fake)+ return create+++def text_model(monkeypatch):+ monkeypatch.setattr(llm, "TEXT_MODEL_NAME", "qwen3")+ monkeypatch.setattr(llm, "TEXT_MODEL_BASE_URL", "http://vllm:8000/v1")+ monkeypatch.setattr(llm, "TEXT_MODEL_API_KEY", None)+ monkeypatch.setattr(llm, "CHAT_MODEL_NAME", None)+ monkeypatch.setattr(llm, "CHAT_MODEL_API_KEY", None)+ monkeypatch.setattr(llm, "CHAT_MODEL_BASE_URL", None)+ monkeypatch.delenv("GPT5_FORCE_MODE", raising=False)++[email protected]("raw, expected", [+ ("", ({}, None)),+ ('{"chat_template_kwargs": {"enable_thinking": false}}',+ ({"chat_template_kwargs": {"enable_thinking": False}}, None)),+ ('{"reasoning": {"effort": "low"}}', ({"reasoning": {"effort": "low"}}, None)),+])+def test_parse_valid_settings(monkeypatch, raw, expected):+ monkeypatch.setenv("TEXT_MODEL_EXTRA_BODY", raw)+ assert llm._parse_extra_body("TEXT_MODEL_EXTRA_BODY") == expected++[email protected]("raw, message", [+ ("{reasoning: low}", "is not valid JSON"),+ ('["low"]', "must be a JSON object"),+ ('"low"', "must be a JSON object"),+])+def test_parse_invalid_settings(monkeypatch, raw, message):+ monkeypatch.setenv("TEXT_MODEL_EXTRA_BODY", raw)+ value, error = llm._parse_extra_body("TEXT_MODEL_EXTRA_BODY")+ assert value == {} and message in error and "TEXT_MODEL_EXTRA_BODY" in error+++def test_no_setting_sends_no_extra_body(monkeypatch, text_model):+ create = _client(monkeypatch, "client")+ monkeypatch.setattr(llm, "TEXT_MODEL_EXTRA_BODY", {})+ llm.call_llm_completion([{"role": "user", "content": "q"}])+ assert "extra_body" not in create.call_args.kwargs+++def test_text_model_parameters_are_sent(monkeypatch, text_model):+ create = _client(monkeypatch, "client")+ body = {"chat_template_kwargs": {"enable_thinking": False}}+ monkeypatch.setattr(llm, "TEXT_MODEL_EXTRA_BODY", body)+ llm.call_llm_completion([{"role": "user", "content": "q"}], max_tokens=100)+ kwargs = create.call_args.kwargs+ assert kwargs["extra_body"] == body+ assert kwargs["extra_body"] is not body # a copy per request+ assert kwargs["max_tokens"] == 100+++def test_parameters_are_added_to_gpt5_requests(monkeypatch, text_model):+ create = _client(monkeypatch, "client")+ monkeypatch.setenv("GPT5_FORCE_MODE", "true")+ monkeypatch.setattr(llm, "TEXT_MODEL_EXTRA_BODY", {"reasoning_effort": "minimal"})+ llm.call_llm_completion([{"role": "user", "content": "q"}])+ kwargs = create.call_args.kwargs+ # The SDK merges extra_body last, so the configured value is the one sent.+ assert kwargs["reasoning_effort"] == "medium"+ assert kwargs["extra_body"] == {"reasoning_effort": "minimal"}+++def test_invalid_setting_fails_the_request(monkeypatch, text_model):+ create = _client(monkeypatch, "client")+ monkeypatch.setattr(llm, "TEXT_MODEL_EXTRA_BODY_ERROR", "TEXT_MODEL_EXTRA_BODY is not valid JSON (x)")+ with pytest.raises(ValueError, match="TEXT_MODEL_EXTRA_BODY is not valid JSON"):+ llm.call_llm_completion([{"role": "user", "content": "q"}])+ create.assert_not_called()+++def test_chat_on_the_text_model_uses_its_parameters(monkeypatch, text_model):+ create = _client(monkeypatch, "client", "chat_client")+ monkeypatch.setattr(llm, "TEXT_MODEL_EXTRA_BODY", {"reasoning": {"enabled": False}})+ monkeypatch.setattr(llm, "CHAT_MODEL_EXTRA_BODY", {"ignored": True})+ llm.call_chat_completion([{"role": "user", "content": "q"}])+ assert create.call_args.kwargs["extra_body"] == {"reasoning": {"enabled": False}}+++def test_a_separate_chat_model_does_not_inherit_them(monkeypatch, text_model):+ create = _client(monkeypatch, "client", "chat_client")+ monkeypatch.setattr(llm, "CHAT_MODEL_NAME", "openai/gpt-4o")+ monkeypatch.setattr(llm, "CHAT_MODEL_BASE_URL", "https://openrouter.ai/api/v1")+ monkeypatch.setattr(llm, "TEXT_MODEL_EXTRA_BODY", {"chat_template_kwargs": {"enable_thinking": False}})+ monkeypatch.setattr(llm, "CHAT_MODEL_EXTRA_BODY", {})+ llm.call_chat_completion([{"role": "user", "content": "q"}])+ assert "extra_body" not in create.call_args.kwargs++ monkeypatch.setattr(llm, "CHAT_MODEL_EXTRA_BODY", {"reasoning": {"effort": "low"}})+ llm.call_chat_completion([{"role": "user", "content": "q"}])+ assert create.call_args.kwargs["extra_body"] == {"reasoning": {"effort": "low"}}+++def test_a_rejected_request_names_the_setting(monkeypatch, text_model, caplog):+ response = httpx.Response(400, request=httpx.Request("POST", "http://vllm:8000/v1/chat/completions"))+ error = openai.BadRequestError("unknown field", response=response, body=None)+ _client(monkeypatch, "client", side_effect=error)+ monkeypatch.setattr(llm, "TEXT_MODEL_EXTRA_BODY", {"reasoning": {"effort": "low"}})+ with pytest.raises(openai.BadRequestError):+ llm.call_llm_completion([{"role": "user", "content": "q"}])+ assert "TEXT_MODEL_EXTRA_BODY adds reasoning" in caplog.text+++def test_incognito_requests_use_the_shared_helpers():+ """Incognito title, summary and chat go through call_llm_completion and+ call_chat_completion, so they get the GPT-5 handling and these settings."""+ import ast, os+ root = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))+ for path in ("src/tasks/processing.py", "src/api/recordings.py"):+ tree = ast.parse(open(os.path.join(root, path), encoding="utf-8").read())+ direct = [n.lineno for n in ast.walk(tree) if isinstance(n, ast.Attribute) and n.attr == "create"+ and isinstance(n.value, ast.Attribute) and n.value.attr == "completions"]+ assert direct == [], f"{path} calls the client directly at lines {direct}"diff --git a/tests/test_llm_temperatures.py b/tests/test_llm_temperatures.pyindex d4bf9915..a6483ed5 100644--- a/tests/test_llm_temperatures.py+++ b/tests/test_llm_temperatures.py@@ -142,7 +142,7 @@ def test_incognito_title_and_summary_use_the_resolved_values(ctx):fake_client = MagicMock()fake_client.chat.completions.create.side_effect = fake- with patch.object(processing, 'client', fake_client):+ with patch.object(processing, 'client', fake_client), patch('src.services.llm.client', fake_client):processing._generate_incognito_title('Some transcript text about budgets.')processing.generate_incognito_summary('Some transcript text about budgets.')assert seen == [0.15, 0.0]diff --git a/tests/test_title_prompt.py b/tests/test_title_prompt.pyindex c5b89413..793de0e2 100644--- a/tests/test_title_prompt.py+++ b/tests/test_title_prompt.py@@ -201,7 +201,7 @@ def test_incognito_titles_use_the_user_level(ctx):fake_client = MagicMock()fake_client.chat.completions.create.side_effect = fake_create- with patch("src.tasks.processing.client", new=fake_client):+ with patch("src.tasks.processing.client", new=fake_client), patch("src.services.llm.client", new=fake_client):processing._generate_incognito_title(TRANSCRIPT, user=user)assert '- Start with the date' in calls[-1][1]["content"]