Bring your own prompts
Write and edit your prompts through MCP. Organise them into categories and save a version to run. If you want help building a prompt bank from a question, generation is optional.
Your agent.
Our harness.
Run model experiments from your own agent through MCP. Bring your prompts, choose the models and search modes, and use our tools to configure, launch and inspect each run. Our system handles execution and keeps every answer, transcript and cost. You decide what the results mean.
Join the waitlist
Private beta. Join for access to the harness and MCP tools.
Your prompts, across the models and modes you choose.
- mentioned
- not mentioned
- never measured
- Colour says which mode answered.
This example shows a mention analysis you could build from the saved answers. Each square represents one prompt on one model in one mode. Our system preserves the individual responses so your agent can analyse them on your terms.
Run it through MCP
Give your agent the tools to choose models and search modes, set a budget, start a run, check progress and stop it. You control the experiment; our system runs the harness.
Keep the evidence
Retrieve individual answers, transcripts and costs through MCP. Search saved answers and use them in your own analysis. Failed or unrun items keep their status, so missing evidence stays distinct from a result.