Cinchor – Control what an AI agent can do, and prove what it did
Cinchor provides accountability infrastructure for autonomous agents, using a 'bound before, proven after' mechanism to set limits before an agent acts and generate tamper-proof, independently verifiable records afterward, suitable for regulators, auditors, insurers, or courts. It offers a managed gateway and embedded SDKs.
Accountability infrastructure for autonomous agents
Bound before. Proven after.
Bound what an agent can do before it acts — and prove what it did after, verifiably enough for a regulator, auditor, insurer, or court.
Try it live — no key → Request access
Run the full loop against the live network in your browser — no signup, no key. A key is for building with the SDK against your own tenant.
Auth0 for auth · Datadog for observability · Vanta for compliance → Cinchor for agent accountability.
cinchor · decision stream LIVE
Illustrative. Every line is an on-chain, independently verifiable record.
Two verbs. That's the whole surface.
Authorize-or-refuse a consequential action against a pre-scoped capability — a ceiling, a window, an allowlist. The substrate is the enforcement point: an out-of-scope action commits no state change, no matter how the agent reasons, is prompted, or is compromised.
Commit a tamper-evident, independently verifiable record of a decision and the full context behind it, bound to the policy in force at the time. Anyone can verify it later without trusting the operator.
Together they turn unbounded irreversible harm into bounded irreversible harm — with a record an adversary can verify.
Two ways to put an agent on the record.
Managed gateway
Early access
An API key and HTTPS. No wallet, no gas, no RPC, no node. The fastest path from "my agent acts" to "my agent is accountable."
Issue capabilities, enforce, attest, and query the audit trail over REST.
Cinchor runs the keys, the gas, and the substrate connection.
One integration; accountability as a service.
Request access →
Embedded SDKs
Available now
For teams comfortable holding a key. Import a library, wrap your agent's decision and action points, call the two verbs.
@cinchor/sdk — TypeScript · cinchor — Python · Go
Parity-matched: a capability minted in one language verifies in another.
You never "go on a blockchain" — you call two functions.
See the quickstart ↓
Three calls to bound-and-prove.
install
npm install @cinchor/sdk
// 1 · a principal mints a scoped capability to an agent const { capabilityId } = await cinchor.mintCapability({ principal, // the granting party (a signer) agent: agentAddress, maxSpend: 100n, ttlSeconds: 3600, });
// 2 · the agent enforces an action — out-of-scope = no state change
const outcome = await cinchor.enforce({ capability: capabilityId, agent, amount: 40n });
if (!outcome.allowed) throw new Error(refused by the substrate: ${outcome.reason});
// 3 · attest the decision — provable after, by anyone const { attestationId } = await cinchor.attest({ capability: capabilityId, agent, context });
// …later, verify without trusting the operator const { ok } = await cinchor.verifyAttestation(context, attestationId);
Same two-verb surface in TypeScript, Python, and Go — each validated end-to-end against a live mesh.
Request access
The managed gateway is in early access. Tell us what your agents do and how they spend, and we'll get you on.
Prefer email? [email protected]