How lyra thinks about verifiable autonomy.
Notes and deep-dives on the ideas behind the product — the safety model, on-chain identity, and the engineering that makes an AI agent safe with money.
Verifiable autonomy: the AI advises, code enforces
The core idea behind lyra — why an autonomous agent should never hold keys, and how four deterministic gates make “wrong model” a non-event for your funds.
ReadThe four-gate write pipeline
Policy → simulate (dry-run) → approve → execute. How every PTB is checked before it broadcasts on Sui and returns a tx digest.
ReadIdentityA Sui address is the agent identity
Each agent is a Sui keypair; its on-chain history of policy-checked PTBs and Walrus receipts lets its track record be checked, not just claimed.
ReadPolicyPolicy as code, not prompts
Why fund controls live in deterministic code and the on-chain lyra::policy Move package — allowlists, caps, slippage and health-factor floors.
ReadRuntimeThe agent’s brain & memory
The OpenAI-compatible brain, local file-based memory, and how context is assembled per turn.
ReadRuntimeTools & plugins
The tool registry and plugin host that give the agent real capabilities — read, swap, lend, transfer.
ReadSourceRead the source
lyra is open source. Browse the policy spine, the lyra::policy Move package, and the plugin host on GitHub.
Read