Agent-readable
audits, signed
on-chain.
When one agent trusts code another agent wrote, something has to vouch for it. TryAnneal audits the contract and signs the verdict on-chain — so trust is verifiable, not assumed.
npx anneal audit ./Vault.sol --network mantleis_this_safe()
The same call any agent makes before composing with unknown code. Read a verdict straight from the on-chain registry on Mantle — pick an example or paste a code hash.
Audit in seconds. Attest forever.
Three frontier models, one ground-truth analyzer, one verdict. From contract to on-chain proof in under sixty seconds.
Submit
Drop a contract or paste a GitHub URL. We pull the source, detect the chain, and queue it for analysis.
Audit
ChainGPT triages. Two Groq critics — Llama-3.3-70B and OpenAI GPT-OSS-120B — argue. Slither and Aderyn hold the ground truth. Only multi-model agreement becomes a finding.
Attest
Verdict signed on Mantle via ERC-8004. Findings sealed to Arweave. Reputation moves on-chain — readable by any agent, forever.
Your protocol will get audited by an autonomous agent in three years.
Today, our agents audit yours.
Four models, one verdict
Two architecturally-distinct Groq critics — Llama-3.3-70B and OpenAI GPT-OSS-120B — argue every finding. Slither and Aderyn hold the line on the static-analysis truth. Nothing ships without multi-model agreement — so single-model hallucinations don't reach you.
Live Mantle gas, in USD
Read straight from the L1Block predeploy and GasPriceOracle. We split L2 execution, L1 blob data, and operator fee per function — priced in dollars, refreshed on every audit.
Verdicts live on-chain
Every audit is signed onto Mantle through ERC-8004 Identity and Reputation. Any protocol, indexer, or wallet can verify an agent's history without asking us — or anyone else.
Findings only you can open
Full findings are encrypted AES-256-GCM, sealed to Arweave, and gated by Lit access conditions. The public sees a verdict hash. Your team sees the report. No third party in between.
What no one else ships.
Slither catches what a parser can catch. Frontier models catch what a parser misses. We run both, reach 2-of-3 consensus, and put the verdict on-chain. No one else ships all three.
TryAnneal | AuditBase | Claude Code | Trail of Bits | |
|---|---|---|---|---|
Catch reentrancy and parser-level bugs | ||||
Catch logic bugs only a model can see | ||||
Get a permanent, verifiable record on-chain | ||||
Build agent reputation other contracts can read | ||||
See per-function gas costs in dollars | ||||
Keep your findings private to you | ||||
Ship in minutes, not weeks | ||||
Run it without procurement or sign-up |
How the engine works.
Input
Contract or GitHub URL, chain auto-detected.
Analyze
ChainGPT pre-screen → Groq critics (Llama-3.3-70B + GPT-OSS-120B) cross-validate → Slither + Aderyn.
Validate
2/3 consensus → confidence → verdict.
Attest
ERC-8004 identity, encrypted findings, reputation ±.
On-chain
Settled on Mantle.
Start auditing
in 10 seconds.
One command. Three models. A signed verdict on Mantle. No login, no waiting list.