Public Beta · Design partners

Run your agents under a Constitution

A GoodHelp Constitution is the rules your agents work by — and a tamper-evident record of what they did, why, and on whose authority. We’re inviting a small group of design partners to run it on their own agents while we harden it together.

What your agents run under

Four parts that work together — the rules on one side, the proof on the other.

Rules you write and can see

Each agent runs under a declarative policy — allow-listed tools, a per-run cost cap, an external-comms allow-list, data-scope predicates — decided server-side before the action runs. You can tell at a glance whether enforcement is live for each agent.

Approvals for risky actions

Sensitive actions wait for a second person to approve. Separation-of-duties means whoever edited an agent can’t approve its next action — enforced fail-closed.

A tamper-evident record

Every action — allowed or denied — is appended to a per-org, hash-chained audit log rooted in a Google Cloud KMS key. Any retroactive edit breaks the chain and is detectable on replay.

Proof you can check yourself

Export a signed evidence pack for any date range and verify it offline against our published key — no trust in our dashboard required.

What a design partner gets

Early access to the beta

Run your own agents under a GoodHelp Constitution while we harden the surface with you.

A direct line to the team

Talk to the people building it. Your edge cases and compliance needs shape what ships next.

Regulated-industry focus

DCAA-aligned controls, a 7-year retention profile, and PII redaction at ledger write time for teams who answer to auditors. (Not DCAA-certified — we map controls to the requirements.)

Onboarding help

We help you author your first policies, wire approvals, and export your first evidence pack.

Straight talk about the beta

This is a public beta, and we’d rather be honest than shiny. Today the Constitution gives you a tamper-evident record of what your agents did, why, and on whose authority, with policy decisions made before each action runs. We are not SOC 2 / FedRAMP / DCAA-certified, and self-serve onboarding isn’t instant yet — that’s exactly why we work hands-on with design partners. See everything we do and don’t claim in the Trust Center.

Talk to us

Prefer email? Reach us at info@goodhelp.ai.