Skip to content

Raw (npm package @tlelabs/raw) is an open-source local coding agent from TLE Labs, released under the MIT license. It is built for models with limited context. You decide what each run can use: an agent selects an ordered list of tools, skills and hooks, a system prompt, and one model connection.

  • Explicit capabilities. Tools, skills, hooks and MCP tools run only when an agent selects them. Nothing is added implicitly.
  • Local and hosted models. Connect a local model server such as Ollama, or a hosted API through one of four wire methods: OpenAI-compatible Chat Completions, OpenAI Responses, Anthropic Messages, or Google Generate Content.
  • Saved sessions. Conversations persist across process restarts and can be resumed from any surface.
  • Extensible parts. Write your own tools, skills, hooks and agents, and package them for another machine.

Raw runs tools with your operating system account’s permissions. The working directory sets where relative paths resolve, but it is not a sandbox. Read Permissions before you run Raw on code you do not trust.

Raw has three surfaces that share the same agents and saved sessions:

  • Terminal. Run one task with raw "task", or open a saved REPL with raw.
  • Dashboard. Run raw dashboard to chat in a local browser app, review approvals and context usage, and edit agents, tools, models and packages.
  • ACP. Run raw --acp --stdio so an IDE or a parent agent can drive sessions over the Agent Client Protocol.