Prove what youragents did.
The system of record for AI agent actions. Every tool call, approval, and authorization scope, captured tamper-evident and mapped to the controls an auditor or examiner will ask about.
Change one byte, break every hash after it.
Every agent action is canonicalized, hashed, and linked to the record before it. Alter anything and the chain stops matching from that point on. Auditors confirm the whole sequence themselves, no trust required.
- RFC 8785 canonicalization
- SHA-256 hash chain
- Public verify endpoint
Prove every action was within authority.
Register each agent with the scopes it is authorized for. Every action is checked against them at ingest and mapped to the controls it supports, so the distance between what an agent did and what you can prove disappears.
- Agent registry and scope checks
- HIPAA, NYDFS 500, SEC 17a-4 packs
- SOC 2, ISO 27001, GDPR coverage
A complete record of what every agent did.
First-class agent identity, tool calls, approvals, and human-in-the-loop review. Filter to one agent and one data type to produce the examiner-ready pack that proves it acted within authority.
- Agent identity and tool calls
- Human supervisor links
- Examiner-ready evidence packs
One call records what your agent did.
Send the agent action: a tool call, an approval, a decision. We canonicalize it, hash it, link it to the record before it, and check it against the agent’s authorized scopes, then return a proof anyone can verify later. Send the same action twice and you still get exactly one sealed record.
- Idempotent by header, safe to retry
- Scope-checked against the agent registry
- Independently verifiable by any auditor
await stonewrit.record({event_type: "agent.tool_called",actor: { type: "ai_agent", id: "agt_01HX" },action: { name: "sql.query", category: "data_access" },resource: { type: "db.customers", classification: ["pii"] }}){"id": "evt_01HXZ8K4ME…","sequence": 4218,"hash": { "event": "sha256:7a2c1d…e840" },"chain_position": 4218}Don’t trust us. Verify it yourself.
The mechanism is public. The canonicalization, the hash chain, the verifier, and the event spec are open source under Apache 2.0. Run the verifier against your own evidence and confirm it independently. A black box that asks you to trust it is a weaker claim than code you can read.
Apache 2.0 end to end. Self-host the whole stack.
- Hash-chain coreApache 2.0The canonicalization and hashing, the exact code that seals every event.
- Verifier CLIApache 2.0Recompute and confirm any chain yourself, offline.
- Event specApache 2.0The schema and frozen conformance vectors everything is checked against.
- Baseline classifierApache 2.0The rule engine that maps events to the controls they support.
Priced on evidence, not headcount.
One plan covers your whole organization. Invite the entire team, no per-seat math. You pay for the evidence you record: a generous monthly volume, then transparent metered overage with a spend cap you set. Never a surprise invoice.
Starter
per organization · whole team included
$853 / year (save 10%)
then $0.60 per 1k events over 250k
For teams proving out agent evidence before production.
- 250k events / month included
- Up to 3 projects, unlimited environments
- SOC 2, ISO 27001, HIPAA, GDPR
- JSON evidence exports + hash proofs
- Email support
7-day free trial · card required
Get startedPro
per organization · whole team included
$2,149 / year (save 10%)
then $0.25 per 1k events over 5M
For production agent deployments with compliance requirements.
- 5M events / month included
- Unlimited projects + environments
- SOC 2, ISO 27001, HIPAA, GDPR
- JSON evidence exports + hash proofs
- Google SSO
- Email support, 24h response
- Up to 100 org members
7-day free trial · card required
Get startedEnterprise
per organization · whole team included
For regulated agent fleets needing SSO, retention, or on-prem.
- Unlimited events
- Custom retention up to 7 years
- SAML / OIDC SSO
- Custom + private frameworks
- Dedicated Slack + named CSM
- 99.99% uptime SLA + DPA
- On-prem / VPC deployment
Common questions.
The next time someone asks “can you prove that?”
When legal asks whether your agent stayed within its authority, Stonewrit gives you the answer, provably, instead of hoping your logs hold up.