Five modules, one entry point: observe, validate, execute, protect. A human can always stop the layer.
Five modules that work together. Each one can be inspected separately — but the layer is sold and deployed as a whole.
Every agent action goes through one entry point: auth, stop switch, policy, risk, execution, audit. Fail-closed — if a module errors or times out, the answer is DENY, never "let it through".
Risk engine. Scores every proposed trade against risk rules before execution: position size, exposure, volatility, liquidity, drawdown.
Execution. The only module that holds exchange keys. Applies limits, sends orders, and can cancel everything at once.
On-chain intelligence. Context for every decision: whales, liquidity, clusters, scam signals.
The human in the loop. Instant stop and access revocation — and it holds no exchange keys, so a compromise is a denial of service, not a theft.
Every agent action goes through a single endpoint. The execution layer cannot be switched off by config — it is an invariant, covered by tests.
A halt flag stops everything immediately and independently of other modules. Mass order cancellation is one command away.
Policy and risk produce ALLOW, DENY or ESCALATE. An error or a timeout from any module always becomes DENY.
Execution goes through Palisade; every decision is written to an append-only audit chain whose integrity can be verified on demand.
$99/mo. Sign up, pay and get your keys in the account.