For harness builders and early teams
- Hosted Rivetplane server
- Unlimited connected runners
- REST, WebSocket, and MCP
- Alpha feedback channel
Build or bring the agent harness that fits the work. Rivetplane gives it a common ACP connection, remote API and MCP surface, and one place to operate every session.
Choose the models, tools, memory, permissions, and runtime that fit the job.
Give different harnesses one protocol and one stable control surface.
Use REST, live events, and MCP over the same session state.
The runner dials out. Code, credentials, and permissions stay on the machine.
Rivetplane is not one fixed harness. It is the shared control layer for harnesses you build or already use. The agent loop and permission authority stay with each local harness.
Connect different runners through one open session protocol.
Bind each response to an exact pending ID.
Tail messages, tool calls, results, and state changes.
Select an explicit model reported by that runner.
REST and MCP use the same authenticated operations and session state. There is no private UI-only control path.
List the fleet, load a transcript, inspect the active pending request, then send a message or exact response.
Open API documentationGET /v1/sessions?status=waiting_approval GET /v1/sessions/{id}/pending POST /v1/sessions/{id}/pending/respond { "pending_id": "per_01J...", "response": "approve", "scope": "once" }
The server never opens a connection into a workstation. Model keys, Git credentials, and source files remain local unless a transcript payload explicitly contains them.
Use hosted Rivetplane at no cost while the product is in Alpha. Model and harness-provider charges remain with your existing providers.