A platform for building AI agents that can only do what your rules allow.
Zarel is a platform for building AI agents. Their rules aren't in a prompt the model can ignore, or in code you redeploy every time they change. They live in a contract your team writes. The model proposes each action, but it only runs if the contract allows it: code checks that, not an AI. When a rule changes, your team publishes a new version of the contract, with no redeploy. Every state transition and flow step is hash-chained and sealed with signed checkpoints, so an auditor can check the record offline with a verifier they run themselves. It is tamper-evident: it detects an alteration, it does not prevent one.
Pre-launch. No certifications.
| Package | What it is | License |
|---|---|---|
@zarel-ai/audit-chain |
Offline verifier for the audit hash chain. Zero runtime dependencies. | Apache-2.0 |
@zarel-ai/audit-tsa |
Offline RFC 3161 timestamp-token verifier and Merkle inclusion proofs. | Apache-2.0 |
@zarel-ai/cli |
The zarel CLI, including zarel verify. |
MIT |
@zarel-ai/sdk |
TypeScript SDK for the Zarel API. | MIT |
Their source is public in this organization:
zarel-audit-verifier:audit-chainandaudit-tsa.zarel-cli:cli.zarel-sdk-typescript:sdk.
The runtime itself is closed source.
The reasoning behind the design is at zarel.ai/blog, starting with You can't fight probabilism with probabilism.
hello@zarel.ai · LinkedIn · Security reports: see SECURITY.md.