An LLM is good at deciding what to do and bad at being a safety boundary. So Lyra keeps the model in an advisory seat: it reads intent, picks tools, weighs tradeoffs, and discovers opportunities. It never gets the final word over your funds.
Every value-moving action is checked by a pure, unit-tested policy engine first. Hard caps on amounts, recipient and token allowlists, slippage caps, an autonomy tier. No network, no model. A violation blocks the action outright.
A proposed transaction is dry-run with estimateGas and simulateContract before a single unit of gas is spent. A revert aborts the action with a decoded reason, so doomed transactions never reach the chain.
The approval floor sits beneath the session mode. A material-risk action prompts for human approval even when autonomy is wide open, and is denied outright under strict. Fund controls live in code, not in a prompt the model could talk its way past.
Cleared actions broadcast on Sui, wait for the receipt, and return a decision record: the policy verdict, the simulated gas, and the transaction digest. Every step is reconstructable after the fact.
Balances and tokens, transfers and SUI moves, trading via Cetus and DeepBook, lending via NAVI and Scallop, and yield discovery via DeFiLlama with risk and RWA-eligibility flags. From the terminal, Telegram, or the web console.
Chat it, run it, or build on it. The policy engine keeps the boundary.
