mandala
A governed place to act.
Mandala is the gate between an agent's intent and its side effects. It is deliberately small, openly reviewed, and honest about what is not built yet.
What it is
A pass gate around action: destructive routes can require a signed pass before dispatch, policy versions are pinned and checked, and every governed chain emits a continuity receipt – so what was permitted and what actually happened stay checkable together. The gate does not hold funds, does not phone home, and runs on the same machine as the work.
Where it stands today
- gate-lite ships as a review snapshot: pass gate on destructive routes (off | optional | required), pinned policy versions, and a declared issuer.
- Chains emit spec 0.3 continuity receipts; resumed chains preserve the stored spec, so a mixed history never silently upgrades. End-to-end verified: a gate-lite bundle verifies TRUSTED under the published Python and Rust verifiers and the hosted API.
- The full test suite runs locally; the receipt format it emits is the same one described on the receipts page.
Evaluation infrastructure
For harnesses that need a place to write and query, the evaluation endpoint exposes an Add / Search / Delete contract over a participant-isolated store (token on request; eval-only, no SLA, keyless health check). It exists so evaluations – including the lab's own reliability audits – run against declared limits instead of ad-hoc scripts. Access and limits are described on the hosted lanes page.
Honest deferral
gate-hard – a microVM floor under the gate – remains deferred: the isolation story is not good enough yet to claim, so it is not claimed. gate-lite is a review snapshot, not a security boundary; treat it as a governed workflow with receipts, and read the source before trusting it with anything you would not put in a normal process.