ThinkerWave

An autonomous research agent for hard, open-ended questions. It works the problem round after round, runs its own analysis code, checks its own work, and labels every claim in the answer verified, assumption or guess.

Try it free at thinkerwave.ai GitHub: SDK, CLI, MCP server
A real two-round ThinkerWave run in the terminal, ending with an answer whose claims are labelled verified, assumption or guess

A real run from the command line: the rounds as they happen, then the answer with its evidence labels.

How it works

  1. Describe a hard problem. Paste your numbers and files; they're kept exactly as you give them.
  2. Approve the plan. See what it will aim for. Nothing runs until you press Launch.
  3. It works on its own. It tries, checks and improves, round after round, and keeps the best answer.

In the app

The plan ThinkerWave proposes before a run, with the user's numbers listed as fixed inputs

You approve the plan. Your numbers stay fixed.

A live run: the engine view and the current answer

Then it works on its own, round after round.

From your code or your coding assistant

pip install thinkerwave
tw run "Should we raise prices 15%, 8%, or hold?" --given "1,200 customers" --watch

# Claude Code or Cursor (MCP)
claude mcp add thinkerwave --env THINKERWAVE_API_KEY=tw_live_… -- uvx thinkerwave-mcp

Get an API key at thinkerwave.ai → Settings → API keys. Free runs are included with every account.