Weave is a desktop LLM application for Linux (with Weave Lite for macOS). At its core it's a fast, familiar multi-model chat client — but it ships with a full suite of tooling built around that core: Weaves, sandboxed tool use, MCP support, evaluations, skills, and more.
The signature feature is Weaves: describe one large change, update, or task and Weave breaks it into a collection of small, isolated tasks an agent runs for you. Because context is not shared between tasks, each one is a self-contained logical change — and that's what lets a modest, locally-hosted model do work you'd normally hand to an expensive frontier model.
What you can do with it
- Chat with any model from OpenAI, Anthropic, or a local Ollama server — organized into folders, with full history and token tracking. See Chat.
- Open a project to give the model sandboxed tools that read, edit, and run code inside a directory you choose. See Projects.
- Build Weaves — decompose a big job into isolated tasks and run them independently. See Weaves.
- Stay in control — on Linux every tool call is sandboxed and writes can be reviewed before they touch your files. See Sandboxing.
- Extend and integrate — connect MCP servers for more tools, or drive Weave from other apps over its HTTP API or MCP server.
New here? Start with Installation and the Quickstart.
Hit a snag or have a question? See Troubleshooting, create a Gitlab issue, or email us at [email protected].
