add-ollama-tool

To enable local Ollama model inference, you add a local Ollama MCP server to the container agent. This exposes installed models via stdio MCP, allowing the agent to call them as tools for generating responses using host-local inference.

Using local Ollama tools reduces latency and cost by avoiding routing every task through remote APIs. It enables host-local inference for practical workflows like summarization, translation, and general question answering without external calls.

You configure an MCP server by wiring MCP server registration and tool definitions in the agent-runner. This provides the `ollama_list_models` tool to enumerate installed models and `ollama_generate` to run prompts through a chosen model.

Yes, you can use Ollama with a containerized agent for text summarization and translation. The integration exposes local models as tools via stdio MCP, enabling the agent to perform these practical workflows using host-local inference.

Yes, you need to update per-group cached runner sources when adding an Ollama MCP server. This step is part of wiring the MCP server registration and tool definitions, and must be validated with a clean build.

If the container agent still routes to remote APIs, the MCP server registration may be incomplete. You must ensure the agent-runner is updated, per-group cached runner sources are refreshed, and the integration is validated with a clean build.