Trust the actions, not just the answers.
Why we're building this
For a decade, the hard question about software was whether it was allowed to act and whether you could see what it did. Authorization and observability answered those well. AI agents break the assumption underneath both: an agent can be fully authorized, produce a clean trace, pass every eval, and still leave the real world in the wrong state. The refund never settled. The retry duplicated the charge. The ticket closed before payment confirmed.
That mismatch is the completion gap, and it is where agent trust quietly breaks. It is widest exactly where it hurts most: the actions that touch money and customer trust. A wrong “done” there isn’t a logging problem, it’s a chargeback, a reopened ticket, and a broken promise, all created in a single confident step.
The missing primitive: outcome verification
Closing that gap needs something that didn’t exist before: deterministic verification that a claimed action actually reached the required final state in the system of record. That practice is Agent Action Integrity, and the artifact it produces is a Proof-of-Completion. Every high-risk action gets checked against the source of truth and sealed into a signed receipt, so completion can finally be measured, with a single number: Verified Completion Rate.
How we build
These principles shape every decision:
Read-only by default
Postcept verifies, it doesn't execute. It never needs the access required to move money or change a record.
Deterministic, not another model
No LLM decides the verdict. Postcept checks the system of record and reports what it says.
Proof over claims
Every verification leaves a signed, tamper-evident receipt. “Done” should be evidence, not a promise.
Least privilege by design
Scoped, revocable credentials: the minimum access a postcondition requires, and nothing more.
Where it stands
The /v1 API contract is versioned and stable, and the relay and the receipt verifier are open source, so every “verified” can be checked against the published keys. We start with refunds, credits, and cancellations, where false completion creates the clearest customer, financial, and compliance risk, and we build hands-on with the teams putting high-risk agents into production. If your agents say “done,” but your team still needs proof, we want to hear from you.
Built in the open.
We're building outcome verification with the teams running high-risk agents today. Tell us what you're shipping.