Tempest

Tempest runs coding agents on your machine. Pick any provider, give each subagent its own tools and model, and set permissions per session.

• Any provider, any model

  Use OpenAI, Anthropic, Meta, OpenRouter, or any provider you have a key for. Swap models without changing anything else.

• Subagents

  The main agent can start other agents for a job. Each one runs on its own, and you can watch what it does.

• You pick the tools

  A subagent gets only the tools you give it: read and search, write and shell, or MCP.

• You pick the model

  One subagent can be cheap and fast. Another can be a stronger model. They do not have to match the main chat.

• You pick who can change that

  Lock the tools and the model, or let the main agent override them for one run.

• Permissions per session

  Manual, allow writes, auto, or YOLO. Auto can send risky steps to a smaller model first. Plan and explore stay read-only.

• Trackable todos

  The agent writes a checklist. You see it in the sidebar and in the chat, with what is done and what is in progress.

• Ask-question tool

  The agent can stop and ask you. A subagent can ask the main agent or you. It waits for the answer.

• Clean UI

  Three panes: chats, the transcript, and the diff. Little chrome.

• Extensible

  Save subagent setups as profiles. Add MCP servers. Add providers.