Make one agent workflow able to prove what it did.
MnemoPay is a free, open-source SDK for tamper-evident audit logs, policy checks and human approvals around AI agent actions. The paid part is us working with you on one real workflow, so it ends with something your reviewer can verify rather than a slide.
What a pilot is
We list the actions your agent takes that someone could later ask about (payments, refunds, deploys, data access) and decide which events, rules and approvals matter.
We integrate the SDK or its MCP tools into your agent: events recorded, risky actions gated by policy, a human approval step where you want one.
You get an exported bundle and a verifier your reviewer can run offline, plus a written list of what the evidence shows and does not show.
What you get, and what you do not
- Working code in your repo, under the SDK's Apache-2.0 licence. No lock-in to a hosted service.
- A bundle in an Article 12-shaped layout (mission, events, CSV, chain, checksummed manifest) if that is useful to you. Download a sample bundle built from a synthetic workflow.
- A "limits" note, written with you: for example that the audit chain proves internal consistency, not the time of creation, unless its root is stored or anchored elsewhere.
Not included: legal advice, a conformity assessment, a compliance certificate, or a claim that your system complies with the EU AI Act. Whether your system is high-risk, and whether the evidence is enough, is for you and your counsel. No production MnemoPay deployment anchors on a public chain yet, and live payment settlement has been verified only against mocks and an opt-in test-mode test, so a pilot that moves real money starts with a test-mode phase.
Who it suits
Teams running an agent that takes actions with consequences (money, infrastructure, customer data), in TypeScript or via MCP, who want evidence and approvals more than a dashboard. You should be able to name one workflow, one engineer who can merge changes, and one person who would review the evidence.
In return we ask for candid feedback and, only if you are comfortable and in your words, permission to describe the work by name afterwards. We will not name you otherwise.
Terms
Scope, timeline and price are agreed in writing by email before any work starts. Larger custom pilots are listed on the pricing page; a first pilot can be smaller than that. Seats are limited because the team is small.
Or check readiness yourself (free)
Six questions, scored in your browser. Nothing is sent anywhere unless you click the email button.