Docent is a local terminal AI assistant for files, web search, and MCP

Summary

Docent is an open-source terminal AI assistant designed to run with a local agent on the user's machine. It can chat with PDFs and Office files, search the web, connect MCP servers, manage projects, and preserve sources and the answering model in chat results. When no agent is available, the CLI downloads the SLP framework and required models, starts the agent on a locally selected port, and stores its connection data under the user's Smartloop directory. The first-use setup downloads the SLP framework, a bge-m3 embedding model of about 417 MB for document search, and a default sl-mini base model of about 769 MB, then creates a project and loads the model. The CLI is distributed through prebuilt binaries for Linux, macOS, and Windows, or can be installed from source with Rust. Users can sign in through a browser or token, create and import projects, enable or disable additional models, and inspect agent status, ports, processes, memory, and loaded models. Interactive terminal chat supports web and document progress, file attachments, resumable sessions, plain line-by-line operation, and model or search controls. The project is released under the MIT License.