MCP
Agents create plans, update tasks, stream output, and search institutional memory from Cursor and other MCP clients.
OpenThrottle|AI
The self-hostable harness that turns plans into agent runs and keeps every commit linked to the work that produced it.
Why it exists
One substrate for planning and agentic execution.
Chat tabs forget. Markdown plans drift. Issue trackers never see the agent run. OpenThrottle keeps plans, tasks, output, and commits in one Postgres-backed system — reachable from your editor over MCP, and owned entirely by you.
How it works
Idea in. Traceable work out.
Ralph drives one task at a time. Progress streams live. Merged commits carry Plan-Id and Task-Id so you can ask what shipped — and why the code exists.
- 01
Plan
Capture intent as durable plans and tasks — not scattered Markdown.
- 02
Execute
Run agents through Ralph, one scoped task at a time, with live output.
- 03
Remember
Search plans, tasks, and docs by meaning with semantic memory.
- 04
Trace
Link shipped commits back to the plan and task that produced them.
Surfaces
Same source of truth, wherever you work.
Bring your agent host and editor. Every surface reads and writes the same plans and tasks.
Developer app
A dashboard for humans: what’s in progress, what shipped, and what’s next — without leaving the loop.
Ralph loop
Queue-backed agentic execution with worktrees and commit traceability built for one-task-at-a-time runs.
Self-host
Run it on your box.
Plans, semantic search, and agent tooling—without vendor lock-in. Clone the repo, run it locally, and own your planning knowledge base from day one.
# Clone OpenThrottlegit clone https://github.com/openthrottle/monorepo.git # CD in and get setupcd monorepo && ./scripts/setup.sh # And start moving faster!pnpm run start # developer → :6020# server → :6021