microsoft/autogen · error · ValueError

default_chat_model not found in config.models

Error message

default_chat_model not found in config.models

What it means

GraphRAG LocalSearchTool.from_settings loads the same settings.yaml and requires a chat model registered under the reserved ID 'default_chat_model' in config.models. If missing, ValueError('default_chat_model not found in config.models') is raised before any model or token encoder is created. Local search uses this model for answer synthesis over community/local context.

Source

Thrown at python/packages/autogen-ext/src/autogen_ext/tools/graphrag/_local_search.py:209

    def from_settings(cls, root_dir: Path, config_filepath: Path | None = None) -> "LocalSearchTool":
        """Create a LocalSearchTool instance from GraphRAG settings file.

        Args:
            root_dir: Path to the GraphRAG root directory
            config_filepath: Path to the GraphRAG settings file (optional)

        Returns:
            An initialized LocalSearchTool instance
        """
        # Load GraphRAG config
        config = load_config(root_dir=root_dir, config_filepath=config_filepath)

        # Get the language model configurations from the models section
        chat_model_config = config.models.get(defs.DEFAULT_CHAT_MODEL_ID)
        embedding_model_config = config.models.get(defs.DEFAULT_EMBEDDING_MODEL_ID)

        if chat_model_config is None:
            raise ValueError("default_chat_model not found in config.models")
        if embedding_model_config is None:
            raise ValueError("default_embedding_model not found in config.models")

        # Initialize token encoder based on the model being used
        try:
            token_encoder = tiktoken.encoding_for_model(chat_model_config.model)
        except KeyError:
            # Fallback to cl100k_base if model is not recognized by tiktoken
            token_encoder = tiktoken.get_encoding("cl100k_base")

        # Create the models using ModelManager
        model = ModelManager().get_or_create_chat_model(
            name="local_search_model",
            model_type=chat_model_config.type,
            config=chat_model_config,
        )

        embedder = ModelManager().get_or_create_embedding_model(

View on GitHub (pinned to 027ecf0a37)

Solutions

1. Add a models.default_chat_model entry (type, model, api_key) to settings.yaml.
2. Add models.default_embedding_model as well, since local search raises on it next if absent.
3. Re-generate settings.yaml with the version of the graphrag tooling matching this autogen-ext release.
4. Verify root_dir points at the workspace containing the settings file you edited.

Example fix

# settings.yaml — before
models:
  default_embedding_model:
    type: openai_embedding
    model: text-embedding-3-small

# settings.yaml — after
models:
  default_chat_model:
    type: openai_chat
    model: gpt-4o
    api_key: ${GRAPHRAG_API_KEY}
  default_embedding_model:
    type: openai_embedding
    model: text-embedding-3-small
    api_key: ${GRAPHRAG_API_KEY}

cfg = load_config(root_dir=root_dir, config_filepath=config_filepath)
missing = [k for k in ("default_chat_model",) if k not in cfg.models]
if missing:
    raise ValueError(f"settings.yaml missing models entries: {missing}")

try:
    tool = await LocalSearchTool.from_settings(root_dir=root_dir)
except ValueError as e:
    raise ConfigError(f"GraphRAG config incomplete: {e}") from e

When it happens

Trigger: Calling LocalSearchTool.from_settings(root_dir=..., config_filepath=...) with a settings.yaml whose models dict has no 'default_chat_model' key. Note local search additionally requires 'default_embedding_model'; the chat-model check fires first.

Common situations: Same family of issues as global search: pre-migration settings.yaml, renamed model keys, wrong root_dir, or settings generated by an incompatible graphrag CLI version. Frequently hit when only the embedding model was configured because earlier experiments used only vector lookups.

Related errors

• default_chat_model not found in config.models
• default_embedding_model not found in config.models
• Failed to fetch settings
• Failed to update settings
• modelName is a required property for LMStudioConfig and cann

AI-assisted analysis of microsoft/autogen@027ecf0a37 (2026-08-15). Data as JSON: /api/errors/c675e7636b158f15. Report an issue: GitHub.