aomi is that layer, already built. Your users type plain English; the runtime builds the exact transaction and rehearses it on a live copy of the chain before anything is signed. The user signs the way they always have, and keys never move. It sits above your existing signer — one integration, not a platform program.
Unedited screen captures of the runtime your product would embed. No amounts, no contract addresses, no bridge named: the user says what they want, and the surface works out the rest — then asks the wallet to sign exactly what it proved.
Nobody needs convincing on the surface itself — Coinbase put agents in the Base app, OKX shipped an agentic wallet, Kraken is rebuilding its consumer app around them, Robinhood extended agentic trading to crypto, MetaMask shipped a developer-facing agent wallet. Users already push billions through plain text boxes. What slips the roadmap item is everything underneath the chat UI.
Simulation, threat scanning, spend caps, protocol allowlists, insured signing — the checklist the first movers set is now the minimum. Ship less and the security review vetoes it; ship it all and it's a platform program.
Forked-state simulation infrastructure, a policy engine, multi-chain broadcast across EVM + SVM, model evals — 12–18 months of work competing with everything else on the roadmap.
One hallucinated transaction or one injection headline retires the feature. Chat input is a new, drainer-shaped attack surface — language must never be able to cause an unverified state change.
"Most 'AI wallets' listed in 2025 either no longer function, have quietly shut down, or have no meaningful AI features beyond a chat interface."
import { AomiFrame } from "@aomi-labs/widget-lib"; // the full chat surface, above the existing signer — // wallet actions ride the simulate-before-sign pipeline export function Assistant() { return <AomiFrame height="640px" width="100%" />; }
The rails are not a concept. The same runtime runs in production today — Somm Finance ships a conversational transaction surface on aomi, with app-level fees priced and settled onchain. The wallet surface above is that runtime, white-labeled.
in production · announced on @aomi_labsThe architecture, the pipeline it enforces, the controls around it, and the terms. Nothing here is bolted on — each is a property of a runtime built directly on a full node that forks every chain in ~200ms, EVM + SVM.
Intent parses into the exact transaction — routed through the aggregators already integrated, batched into a single operation.
intent → exact txExecuted against forked, live chain state to prove the real outcome — fill, slippage, balance changes — before anything is signed.
forked state · ~200msSigned by the user's own wallet through the product's existing signer — per-transaction approval, or scoped delegation within caps.
existing signer · unchangedConfirmed onchain with account abstraction and Solana Jito built in — no execution infrastructure for the product team to run.
AA + Solana JitoSpend caps, venue allowlists, per-session limits, per-wallet rules — defined by the product, enforced on every action. Per-transaction human approval by default; scoped delegation where the product allows it.
Chat input is untrusted by design: language can only propose; the pipeline verifies. A failed simulation or out-of-policy proposal stops before signature. Halt an agent instantly, revoke a grant instantly.
Non-custodial by construction. The harness consumes the existing signer interface — an in-house adapter, Para, or wagmi connectors — and key material never moves. The self-custody story the product sells stays exactly true.
Per-app telemetry on every turn; each executed transaction traceable to the user intent that authorized it. Execution-only by construction — no recommendations, no discretion — with logs built for compliance review.
The invariant: aomi can compose and simulate any transaction, but can only ever propose it. Authority to move value remains with the user's key and the product's policy — the property that keeps a conversational surface on the tooling side of the advice line.
The economics follow the convention wallets already accept for embedded execution: the product keeps its take rate; the platform earns a share of incremental executed flow. A conversational surface converts passive balances into transaction volume the current interface never captures — growth in volume per MAU, on the existing fee.
A native-brand conversational trading surface, above the existing signer, gated by simulation and policy — a sprint to embed, not a platform program.
Read the integration docs →