datasette-llm version 0.1a11 introduces support for OpenAI-compatible endpoints, enabling the Datasette plugin to connect with any local or remote LLM service adhering to that protocol. This update allows users to configure a Datasette Agent instance to run local models hosted via LM Studio, Ollama, or llama.cpp without requiring cloud APIs. The release documentation demonstrates this setup using `uvx` to launch a server where the `lmstudio-qwen` model queries a local database to count repositories and releases. The interface displays standard Datasette navigation alongside a chat window that shows tool calls for database listing and SQL execution. Results appear directly in the chat stream as a table with row counts for specific database tables. This capability bridges the gap between static data exploration and generative reasoning by keeping the entire stack on a single machine. Developers can now run the tool locally using standard shell commands without managing separate container instances for the model and the application. The configuration file remains a simple JSON object defining the model name, version, and API base URL.
- Local execution requires no cloud API keys or external internet connectivity.
- Support extends to any service implementing the OpenAI pseudo-standard protocol.
- Configuration is managed via a JSON object passed to the `datasette` command.