feat(mistral): make Mistral SDK base URL configurable

Add a single optional `MISTRAL_BASE_URL` setting that overrides the
`server_url` passed to every `Mistral(...)` SDK client constructed in
the backend: the dense embedder, the OCR client, and the chat LLM
adapter.

When unset (the default), the SDK continues to use its built-in
`https://api.mistral.ai` endpoint, so behavior is unchanged for
existing deployments. When set, requests are routed to the override
URL — useful for hitting a self-hosted Mistral deployment or a
corporate proxy.

Also expose the new setting in `.env.example` and update the
`_create_dense_embedder` factory and the `MistralDenseEmbedder` ctor
to forward the value.

Assisted-by: Mistral-Vibe:foundry-opus-4.7
Generated by Mistral Vibe.
Co-Authored-By: Mistral Vibe <vibe@mistral.ai>

Author: bricecarpentier

.env.example

@@ -47,6 +47,10 @@ SPARSE_EMBEDDER=fastembed_bm25
 OPENAI_API_KEY=
 ANTHROPIC_API_KEY=
 MISTRAL_API_KEY=
+# Optional — override the base URL used by every Mistral client (embedder,
+# OCR, LLM). Unset to use the default (https://api.mistral.ai). Set this to
+# point at a self-hosted Mistral endpoint or a corporate proxy.
+# MISTRAL_BASE_URL=
 
 # Development Settings
 LOCAL_DEVELOPMENT=false

backend/airweave/adapters/llm/mistral.py

@@ -48,7 +48,10 @@ def __init__(
             )
 
         try:
-            self._client = Mistral(api_key=api_key)
+            self._client = Mistral(
+                api_key=api_key,
+                server_url=settings.MISTRAL_BASE_URL,
+            )
         except Exception as e:
             raise RuntimeError(f"Failed to initialize Mistral client: {e}") from e
 

backend/airweave/core/config/settings.py

@@ -74,6 +74,10 @@ class Settings(BaseSettings):
         TEXT2VEC_INFERENCE_URL (str): The URL for text2vec-transformers inference service.
         OPENAI_API_KEY (Optional[str]): The OpenAI API key.
         MISTRAL_API_KEY (Optional[str]): The Mistral AI API key.
+        MISTRAL_BASE_URL (Optional[str]): Override the base URL used by every
+            Mistral SDK client (embeddings, OCR, chat). Unset → use the SDK
+            default (``https://api.mistral.ai``). Useful for routing to a
+            self-hosted Mistral endpoint or a corporate proxy.
         EMBEDDING_DIMENSIONS (int): Embedding dimensions for the stack (provider, Vespa).
         FIRECRAWL_API_KEY (Optional[str]): The FireCrawl API key.
         TEMPORAL_HOST (str): The host of the Temporal server.
@@ -200,6 +204,9 @@ class Settings(BaseSettings):
     OPENAI_API_KEY: Optional[str] = None
     ANTHROPIC_API_KEY: Optional[str] = None
     MISTRAL_API_KEY: Optional[str] = None
+    # Optional override for the Mistral SDK base URL. When set, applied to all
+    # Mistral clients (embedder, OCR, LLM). Unset → SDK default.
+    MISTRAL_BASE_URL: Optional[str] = None
     FIRECRAWL_API_KEY: Optional[str] = None
     GROQ_API_KEY: Optional[str] = None
     COHERE_API_KEY: Optional[str] = None

backend/airweave/core/container/factory.py

@@ -809,6 +809,7 @@ def _create_dense_embedder(
             api_key=settings.MISTRAL_API_KEY,
             model=spec.api_model_name,
             dimensions=EMBEDDING_DIMENSIONS,
+            server_url=settings.MISTRAL_BASE_URL,
         )
 
     if spec.embedder_class_ref is LocalDenseEmbedder:

backend/airweave/domains/embedders/dense/mistral.py

@@ -45,17 +45,20 @@ def __init__(
         api_key: str,
         model: str,
         dimensions: int,
+        server_url: str | None = None,
     ) -> None:
         """Initialize the Mistral dense embedder.
 
         Args:
             api_key: Mistral API key.
             model: Model name (e.g. "mistral-embed").
             dimensions: Expected output dimensions (validated on response only).
+            server_url: Optional override for the Mistral API base URL. When
+                ``None`` (default), the SDK uses ``https://api.mistral.ai``.
         """
         self._model = model
         self._dimensions = dimensions
-        self._client = Mistral(api_key=api_key)
+        self._client = Mistral(api_key=api_key, server_url=server_url)
         self._tokenizer = get_tokenizer(model)
         self._semaphore = asyncio.Semaphore(self._MAX_CONCURRENT_REQUESTS)
 

backend/airweave/domains/ocr/mistral/ocr_client.py

@@ -86,6 +86,7 @@ def ensure_initialized(self) -> None:
 
             self._client = Mistral(
                 api_key=settings.MISTRAL_API_KEY,
+                server_url=settings.MISTRAL_BASE_URL,
                 timeout_ms=120_000,  # 2 minutes per OCR call
             )
             self._initialized = True

backend/tests/unit/core/test_dense_embedder_factory.py

@@ -18,6 +18,7 @@ def _make_settings(**overrides):
     defaults = dict(
         OPENAI_API_KEY="sk-test-openai",
         MISTRAL_API_KEY="test-mistral-key",
+        MISTRAL_BASE_URL=None,
         TEXT2VEC_INFERENCE_URL="http://localhost:9878",
     )
     defaults.update(overrides)