Prove everything your agents do. Gate the things they should not.
The Agent Action Ledger gives every action your agents take a gate it cannot bypass and a signed receipt you can verify offline. Hunta Gather remembers. This proves.
Memory you can trust · Actions you can prove · One console, one set of signing keys
The problem
Agents act. Almost nothing records it, and nothing governs it.
A production agent calls tools, writes files, moves money, and talks to your customers. The recorders that exist today watch after the fact, for debugging. That leaves three gaps.
Actions live in scattered, mutable logs. Before a regulator, an insurer, or your own incident review, a log anyone could have edited is worth nothing.
A wire transfer, a production delete, an email to a customer. By the time it shows up in observability, it already happened. There was no gate.
Observability tools watch for debugging. Governance, risk and compliance (GRC) checklists are static. Neither produces a tamper-evident, replayable account of who authorized what.
The proposed solution
An approval-gated action ledger
Not another passive flight recorder. Recorders tell you what happened. A gate stops what should not. The Ledger does both, and it reuses the exact machinery Gather already ships.
- 01Gate
Every risky action passes a policy your agent cannot bypass: allow, deny, or route to a human for approval. It is the same review gate Hunta Gather already runs on memory writes, pointed at actions instead.
- 02Sign
Every decision (allowed, denied, overridden) is appended to a per-tenant hash chain and signed with Ed25519, a standard public-key signature anyone can check. Tamper-evident by construction, not by policy.
- 03Verify
Export the chain and verify it offline against our public key, any time, long after the fact. It is the same signed attestation, and the same offline check, your security team already runs on Hunta Gather's tenant isolation.
Other tools record actions and hand you an export. The Ledger puts a gate before the action and a signature after it, from the same console and the same keys as your Gather memory. The property attackers and mistakes cannot get around: the agent proposes, a policy disposes, and the receipt is signed either way.
Honest status
This does not exist yet. We are building it with the first people who need it.
Agent-liability insurers now want a structured evidence trail before they will quote, and the standards bodies are converging on signed receipts. We would rather build the Ledger around a real use case than guess. Tell us yours and we will build it with you.
Join the waitlist
Send us the action you most need gated and proven. Early use cases shape what ships first.