live on monad testnet

Every decision your trading agent makes, committed onchain before it executes.

See it. Verify it. Trust it.

Audit trails you don't have to trust

AI trading agents already produce audit trails -- but they're self-declared and editable by the same party being audited. GlassBox makes tampering visible instead of asking you to take our word for it.

Propose

A trading agent proposes an action. An independent risk agent evaluates it against policy -- without seeing the trading agent's reasoning.

Commit

Both the proposal and the verdict are hashed and committed to an append-only contract on Monad testnet -- before execution is allowed to proceed.

Verify

Anyone can recompute the hash from the raw record and compare it to what's onchain. No trust in our frontend or database required.