Skip to content

Speakr is a personal, self-hosted web application designed for transcribing audio recordings

Python

No language model client unless one is configured (#403 follow-up)

Murtaza Nasir192c878052d3
diff --git a/src/config/app_config.py b/src/config/app_config.py
index 821ce331..af8c3f4d 100644
--- a/src/config/app_config.py
+++ b/src/config/app_config.py
@@ -155,8 +155,10 @@ def initialize_config(app):
client = None
try:
+ from src.services.llm import llm_timeout, LLM_MAX_RETRIES, TEXT_MODEL_CONFIGURED
+ if not TEXT_MODEL_CONFIGURED:
+ raise RuntimeError("no text model configured (set TEXT_MODEL_BASE_URL, and TEXT_MODEL_API_KEY for a hosted service)")
api_key = TEXT_MODEL_API_KEY or "not-needed"
- from src.services.llm import llm_timeout, LLM_MAX_RETRIES
client = OpenAI(api_key=api_key, base_url=TEXT_MODEL_BASE_URL, http_client=http_client_no_proxy, timeout=llm_timeout, max_retries=LLM_MAX_RETRIES)
app.logger.info(f"LLM client initialized: {TEXT_MODEL_BASE_URL} / {TEXT_MODEL_NAME}")
except Exception as e:
diff --git a/src/services/llm.py b/src/services/llm.py
index 7c4f3a31..1ac19315 100644
--- a/src/services/llm.py
+++ b/src/services/llm.py
@@ -153,8 +153,15 @@ http_client_no_proxy = httpx.Client(
headers=app_headers
)
-# Create client with placeholder key if not provided (allows app to start)
+# A language model is configured when it has a key or its server address was
+# set explicitly; a local server often needs no key (#403). Without either,
+# no client is created and nothing is sent: the default address is a hosted
+# service, and a transcript must never go there unasked.
+TEXT_MODEL_CONFIGURED = bool(TEXT_MODEL_API_KEY) or bool(os.environ.get("TEXT_MODEL_BASE_URL", "").split('#')[0].strip())
+
try:
+ if not TEXT_MODEL_CONFIGURED:
+ raise RuntimeError("no text model configured (set TEXT_MODEL_BASE_URL, and TEXT_MODEL_API_KEY for a hosted service)")
api_key = TEXT_MODEL_API_KEY or "not-needed"
client = OpenAI(
api_key=api_key,
@@ -180,6 +187,8 @@ def _build_chat_client():
chat_config = get_chat_config()
if (chat_config['api_key'], chat_config['base_url']) == (TEXT_MODEL_API_KEY, TEXT_MODEL_BASE_URL):
return client
+ if not (CHAT_MODEL_API_KEY or CHAT_MODEL_BASE_URL or TEXT_MODEL_CONFIGURED):
+ return None
dedicated = OpenAI(
api_key=chat_config['api_key'] or "not-needed",
base_url=chat_config['base_url'],
@@ -245,7 +254,7 @@ def call_llm_completion(messages, temperature=0.7, response_format=None, stream=
OpenAI completion object or generator (if streaming)
"""
if not client:
- raise ValueError("LLM client not initialized")
+ raise ValueError("No text model is configured. Set TEXT_MODEL_BASE_URL, and TEXT_MODEL_API_KEY for a hosted service.")
# Check budget before making the call
if user_id and operation_type:
diff --git a/tests/test_cov_llm.py b/tests/test_cov_llm.py
index bfde963f..2535778a 100644
--- a/tests/test_cov_llm.py
+++ b/tests/test_cov_llm.py
@@ -183,7 +183,7 @@ def test_get_chat_config_base_url_falls_back_to_text(monkeypatch):
def test_call_llm_completion_no_client(monkeypatch):
monkeypatch.setattr(llm, "client", None)
- with pytest.raises(ValueError, match="not initialized"):
+ with pytest.raises(ValueError, match="No text model is configured"):
call_llm_completion([{"role": "user", "content": "hi"}])
diff --git a/tests/test_keyless_llm.py b/tests/test_keyless_llm.py
index 27143386..6c3be734 100644
--- a/tests/test_keyless_llm.py
+++ b/tests/test_keyless_llm.py
@@ -1,5 +1,9 @@
"""Language models without an API key (#403).
+With neither a key nor an explicit server address, no model is configured:
+no client is created, so nothing (no transcript) is sent to the default
+hosted address, and model steps report that no model is configured.
+
A local server (Ollama, vLLM, llama.cpp) often runs without a key. Every
model call refused to run when TEXT_MODEL_API_KEY was empty, although the
client was built with a placeholder for exactly this case, and the chat
@@ -74,3 +78,38 @@ def test_a_chat_model_with_its_own_key(env):
def test_a_chat_name_alone_keeps_the_text_model(env):
env(text_key="sk-text", chat_name="chat-model")
assert llm.get_chat_config()["model_name"] == "text-model"
+
+
+_PROBE = (
+ "import json, src.services.llm as l; "
+ "print(json.dumps([l.TEXT_MODEL_CONFIGURED, l.client is not None, l.chat_client is not None]))"
+)
+
+
+def _clients(**env):
+ import json
+ import subprocess
+ root = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
+ clean = {k: v for k, v in os.environ.items()
+ if not k.startswith(("TEXT_MODEL_", "CHAT_MODEL_"))}
+ clean.update(env)
+ out = subprocess.run([sys.executable, "-c", _PROBE], cwd=root, env=clean,
+ capture_output=True, text=True, timeout=120)
+ assert out.returncode == 0, out.stderr[-2000:]
+ return json.loads(out.stdout.strip().splitlines()[-1])
+
+
[email protected]("env,expected", [
+ ({}, [False, False, False]), # nothing set: nothing is sent
+ ({"TEXT_MODEL_BASE_URL": "http://llm.lan/v1"}, [True, True, True]), # local server, no key
+ ({"TEXT_MODEL_API_KEY": "sk-x"}, [True, True, True]), # hosted default with a key
+ ({"CHAT_MODEL_NAME": "c", "CHAT_MODEL_BASE_URL": "http://chat.lan/v1"}, [False, False, True]),
+])
+def test_a_model_is_configured_only_when_set(env, expected):
+ assert _clients(**env) == expected
+
+
+def test_without_a_model_the_call_says_so(monkeypatch):
+ monkeypatch.setattr(llm, "client", None)
+ with pytest.raises(ValueError, match="No text model is configured"):
+ llm.call_llm_completion([{"role": "user", "content": "hi"}])