Run integration tests (requires GOOGLE_API_KEY).

Run unit tests without pytest.

Serialize a list of messages, extracting any system/developer message. Returns: A tuple of (messages, system_message) where syste

Check if compaction should be triggered based on current token usage. Returns: True if token usage exceeds the threshold and

Update the tracked token usage from a response. Args: usage: The usage information from the last LLM response.

Allow this Protocol to be used in Pydantic models. This is useful for type-safe agent settings. Returns a schema that accepts any obj

Lazy import mechanism for heavy chat model imports and model instances.

Create model instances on demand with API keys from environment.

Format token count with k suffix for thousands

Invoke the model with the given messages and optional tools. Args: messages: List of chat messages tools: Op

Invoke the model with the given messages and optional tools. Args: messages: List of chat messages tools: Op

Run a bash command in the sandbox working directory.

Clean up old cache files, keeping only the most recent ones

Clear usage history

Count how many users have a given role.

Create a new message list from a summary. Args: summary: The summary text to use as the new conversation start. Retu

Create Gemini-optimized schema, preserving explicit `required` arrays so Gemini respects mandatory fields defined by the caller.

Create a ToolDefinition from a Pydantic model. This is a convenience method for creating tool definitions from Pydantic models.

Generate the ToolDefinition for this tool.

Call this when the task is finished.

Replace old_string with new_string in a file.

Ensure all objects have additionalProperties: false

Ensure pricing data is loaded in the background. Call this after creating the service.

Flush any pending tool response parts as a single Content message.

Get cost breakdown by model

Provide the database instance.

Get the current status of observability features.

Dependency injection marker. Override this in the agent.

Get usage tokens for a specific model

Fetch a user from the database by ID.

Find files matching a glob pattern like **/*.py

Search for pattern in files recursively.

Check if the response contains tool calls.

Check if we're currently in debug mode.

Check if lmnr is available for tracing.

List all users in the database.

Load message history to continue a previous conversation. Use this to resume a conversation from previously saved state, e.g., when l