a group decision your agents can't lie about, and you can check

Your AI speaks for you. Check what it agreed to.

Several agents, opposite private interests, one decision under rules enforced in code, and a receipt anyone can check. Watch a shortcut that would cheat someone get blocked in code.

Pick a sample scenario below, then hit Run. Same engine, sample inputs.
no crypto · self-hosted you verify locally red lines enforced in code

Sample scenarios cover KYC/AML sign-off, DORA-style approvals and contract negotiation. They are scripted inputs, not reports of use. try the Onboarding scenario.

Three properties · enforced in code

Red lines enforced in code

A hard limit is a code predicate. An option that crosses one is rejected outright and never negotiated away.

Max-min consensus, not majority vote

Among the options every party accepts, the pick lifts the least-happy party. If nothing clears every red line, the result is an honest deadlock.

A receipt you verify yourself

The transcript records masked verdicts only, sealed with SHA-256. Each owner replays their own private limits locally. No sheet is ever revealed.


Live demo

…

The people involved. Each keeps its position private

The options on the table

Each chip is a masked verdict. The coordinator sees only who accepted, never which limit was crossed.

every verdict and hash on this page was computed by the real engine. Reproduce them yourself with docker build -t parley . && docker run --rm -p 8080:8080 parley

The fair answer: lift the least-happy party

The receipt: a tamper-evident, replayable transcript

running…
✓ verified Every party replayed its own limits against the decision; none crossed. Tamper-evident (SHA-256).