← Answers

ICP-1 · Terminal-for-Agents Builder

Does Kestrel train on or capture my model?

No. Brains stay outside: the platform hosts deterministic reflexes and evidence, never an agent’s intelligence. Your model runs on your side and calls the substrate over HTTP (or its equal MCP/CLI/SDK projections). Paid usage is 100% proprietary and never trained on; only anonymous free traces carry a training license, and even then the model itself never enters the platform.

Brains stay outside

This is a platform non-negotiable. Kestrel Markets hosts the deterministic reflexes and evidence — managed data, authority, execution, and proof — while your agent’s intelligence stays external. The platform never sees your weights, your prompts-as-IP, or your policy; it sees the calls your agent makes over the canonical HTTP contract (and its equal MCP / CLI / SDK projections).

The lock-in question

There is no model capture and no lock-in of your brain. Paid usage is 100% proprietary and never trained on. Anonymous free usage grants a training license over its own traces only — never your model — and you can move to paid to make even the traces proprietary. Your intelligence is portable by construction. The training and licensing terms — what a free-tier trace covers, and what paid buys — have their own page: /answers/does-kestrel-train-on-my-strategy.

Reproduce it

The proof is self-verifying — recompute it, do not trust it:

npx kestrel.markets@latest certify <proof-url>