Execute a task by directly querying the model. For models that don't support tools, we'll describe available tools in the prompt.

Check if MCP packages are available.

Run the agent on a specific task. Args: task: The task/query to execute **kwargs: Additional arguments to pa

Return formatted signatures for tools matching *query*. The search is a case-insensitive substring match against both the tool name a

Add a search agent to the orchestrator.

Return whether a model advertises support for an OpenRouter parameter.

Build the contents kwargs for search_and_contents. The Exa API allows multiple content types to be requested simultaneously.

Create a model instance for the given provider. Centralised factory to avoid duplication. Raises: ModelInitializationError: If model

Return a built-in use-case recipe by id.

Determine the provider for a given model ID.

Initialize a model based on model_id and settings configuration. Returns: Initialized model instance or None if initialization fails

Validate the search type, fall back to None (Exa default) if unrecognized.

Try SerpAPI first, fall back to DuckDuckGo.

Validate and normalise a URL. Returns error message or None if valid.

Get recommendations based on past search patterns

Pick the best available content for a short snippet, cascading through fields.

Build a stand-in for an exa_py Result object.

Validate the category, fall back to None if unrecognized.

Shared SerpAPI search logic. Returns formatted results or None on failure.

Add an MCP server configuration. Args: name: Unique identifier for this server server_config: Server configu

Connect to a single MCP server by name. Args: server_name: Name of the server to connect to Yields:

Execute workflow with multi-model coordination and optimization

Get the API key for a specific provider.

Context manager to quickly get MCP tools from a single server. This is the recommended way to use MCP tools as it properly handles the c

Remember successful search patterns for different query types

Initialize data analysis agent with analysis-specific configuration.

Format Exa results matching the existing web_search output style.

Check if MCP packages are available and raise informative error if not.

Execute a search task.

Execute a browser automation task.

Build from an exa_py Result object (or anything with the expected attrs).

Load MCP server configurations from environment variables. Expected format: MCP_SERVER_1_TYPE=stdio MCP_SERVER_1_PATH=/path/

Log search attempts for analysis and optimization

Scrape structured data from a website. Args: url: The URL to scrape data_schema: Schema describing the expec

Create a specialized prompt based on the agent's role.

Get the default set of tools for the agent.

Shared DuckDuckGo search logic.

Create a ``CodeAgent`` that accesses all *tools* through a ``ToolFacade``. Args: tools: Tools to expose through the facade (MCP tools, de

Scrape content from a web page.

Try to get any available model based on settings configuration. Returns: First successfully initialized model or None.

Get performance statistics for this agent.

Check whether the tool can run (SDK installed and API key set).

List all available models based on configured API keys.

Parse CLI worker overrides in the form worker=model.

Conduct research on a specific topic. Args: topic: The research topic depth: Research depth - "quick", "mode

Resolve configured short aliases to provider-ready model IDs.

Resolve bundle names into available tools while reporting missing bundles.

Resolve direct and worker tool assignments for a recipe.

Setup workers with strategic model assignments Args: default_model: Model to use for all workers (if not specified, uses role-bas

Setup workers from a declarative K-LLM use-case recipe.

Return a LiteLLM-compatible OpenRouter model ID.

Update context based on agent results

Initialize the Browser Use tool with API key.

Create an Exa client and tag it with the integration header.

Create server parameters based on transport type. Returns: StdioServerParameters for stdio or dict for HTTP transport

Use the orchestrator agent to synthesize results from multiple agents.

Analyze search performance across engines and query types

Convenience function to create and set up a search orchestrator. Args: tools: List of tools to provide to agents model: Mode

Get the default/fallback model for a provider.

Strategic model assignment based on agent role and capabilities

Get information about a model instance.

Get a performance report for all agents.

Get a summary of current task progress

Get the server parameters for this integration.

List all configured server names.

List built-in use-case recipes in stable order.

Main entry point for the CLI.

Return the OpenRouter API model ID without the LiteLLM ``openrouter/`` prefix.

Prepare context for agent initialization

Return recent models that are especially useful for agent workflows.

Remove a server configuration by name.

Set up a default set of search agents with different specializations.

Return models sorted newest first.

Extract the stable fields the CLI and docs care about.

Return a JSON-serializable summary of a prepared use case.

Temporarily hide the non-serialisable facade during E2B serialisation.

Return resolved tool names for a worker.

Validate configuration at startup and return warnings for missing keys.

Return worker names in execution order.

Write an article on a given topic. Args: topic: The topic to write about style: Writing style (informative,

Initialize the Internet Agent. Args: model_id: The model to use for the agent tools: List of tools to make a

Aggregate results from multiple agents.

Analyze results across different search engines

Check if a model supports code generation for CodeAgent

Check if a model supports tool calling