FIND THE CAUSE.
PROVE THE FIX.
Deterministic tools and a deep AI review go through your EVM-compatible contracts or your DAML source. Every finding is verified before it reaches you, then merged and deduplicated into one report your team can act on — clear, prioritised and ready to fix.
Deterministic tools and a deep AI review read your EVM or DAML source. Every finding verified, one clear report.
Automated — no human reviewer. No audit, human or otherwise, can prove a contract is bug-free.
Every finding earns its place
A claim only reaches you once it has been verified.
Are your AI agents paying in crypto? Audit the account they spend through.
An agent pays on-chain through a smart account and a session key or delegation you granted it. If that grant can be stretched — a bigger spend, a later expiry, a contract outside its scope — the agent, or whoever takes its key, can drain the account. When your code holds ERC-4337, ERC-7579, ERC-7710 or session-key authority, a dedicated delegation review joins the audit and checks exactly that.Agents pay through a smart account and the session key or delegation you gave them. If that grant can be stretched, the account can be drained. Our delegation review checks exactly that.
Every EVM chain
Any EVM-compatible contract — Ethereum, Polygon, BNB Chain, Arbitrum, Optimism, Base. A deterministic tool suite analyses your Solidity in an isolated environment, and everything lands in one report.Any EVM chain, checked by deterministic tools in isolation, delivered as one report.
Deep AI review
Our AI review catches what automated scanners miss — and every finding is verified before it reaches you.
One list, no noise
Every finding is deduplicated into a single list — no repeated claims, no manual reconciling.
Multi-round engagements
Fix, accept risk, or dispute each finding across rounds — the way a real audit firm re-checks work until the case closes.
Dig deeper or dispute
Push back on any AI-sourced finding — ask for a concrete PoC, or argue it’s a false positive — and get a reasoned verdict.
Pay in USDC
One-click MetaMask checkout — no manual copy-paste, no card required.
A review built for DAML
Not an EVM audit with the names changed. Four risk axes, a review written for the language, authorization findings proven by execution, and your upgrade checked against the version it replaces.Built for the language: four risk axes, proven authorization findings, and a checked upgrade.
Verified before it’s a report
A finding only reaches you once it has survived verification. Nothing in your report rests on a single opinion.Nothing in your report rests on a single opinion.
Deterministic tools
A deterministic tool suite runs over your source in an isolated environment.
AI review
A deep review for the failures scanners miss.
Verification
Every potential issue must be verified before it can become a finding.
Final review
Confirmed findings get one last review before they count.
Your report
Everything is merged into one case file.
Open the workspaceDelegated spending gets its own reviewer
Giving an AI agent a wallet means giving it a session key or a delegation on a smart account. That grant can fail in ways an ordinary audit walks straight past: a key that spends more than it was granted, keeps working after it expired, or reaches a contract it was never scoped to. When your code carries an ERC-4337 account, ERC-7579 modules, ERC-7710 delegation or session keys, a dedicated delegation review joins the audit — written for exactly these failures — and its findings are checked like every other.An AI agent spends through a session key or delegation. If that grant can be stretched — more spend, past expiry, outside its scope — the account is at risk. A dedicated review checks exactly that.
A limit checked when a call is validated but not when it executes, or applied to the first call of a batch and not the rest.Checked when a call is validated but not when it runs, or on a batch’s first call only.
Per-key scopes kept where one key can overwrite another’s, or read back from the wrong place when the key signs.One key’s scope can overwrite another’s, or be read back wrong when it signs.
Each re-delegation should only shrink authority. A chain that checks only the last link’s limits lets it quietly widen.A re-delegation that widens authority instead of narrowing it.
A session key that keeps signing after its expiry, or after the owner revoked it.
- Scope reach: every target and function a key was granted, traced to what it can actually call.
- Built-in ERC-4337 knowledge, so the EntryPoint’s own guarantees are never reported back to you as bugs.
- Runs as part of the standard audit, in the same single report.
A DAML audit is not an EVM audit with the names changed
Point an EVM audit at a DAML package and it finds nothing — there is no msg.sender, no gas, no reentrancy, no proxies for it to look at, and the failures that actually cost you here have no EVM equivalent at all. So we built the review DAML needs: four risk axes, a review written for the language, every authorization finding proven by execution, and — when you give us the version you are replacing — a check of your upgrade. Pick the DAML skill in the workspace and this is what runs.An EVM audit finds nothing in DAML — no msg.sender, no gas, no reentrancy. So we built a review for it: four risk axes and every authorization finding proven by execution.
The ledger rejects an under-authorized submission before any code runs, so there is no onlyOwner to forget. The defect is authority wired to the wrong party, which the ledger then enforces faithfully, forever, without ever erroring.The ledger blocks under-authorized commands, so the risk is authority given to the wrong party — then enforced, silently, forever.
The signature risk class here: over-disclosure is only a defect on a shared ledger, and leaking to the wrong participant is frequently worse than an authorization slip.Showing a contract to the wrong participant can be worse than an authorization slip.
A correctness property, not an operational one. One shared contract every participant must consume fails under ordinary honest load — the defect that works with one user and collapses with fifty.One shared contract everyone must consume: fine with one user, fails with fifty.
An axis the EVM has no version of: what your next version would break for contracts already live on the ledger.What your next version would break for contracts already live.
- EVM tooling is built for how Solidity fails, so no third-party analyzer is run on a DAML package: the review below carries it, in minutes, at the same price as an EVM audit.
- A review written for DAML reads your package, and every finding is verified exactly as in an EVM audit.
- Your package must build before the review starts, so every finding is against code that actually compiles.
Everything inside the workspace
Four areas, in the order you’ll use them — the same four you’ll see in the sidebar.
Submit the contract
Upload .sol or .daml files, a whole folder, or fetch a public GitHub repo by URL. Pay in points or with one-click MetaMask.
Tools and AI review it
Deterministic tools and a deep AI review go through your source, and everything lands in one list. Filter by severity, dig deeper on any finding, or dispute it.
Fix, justify, re-check
Mark a finding fixed or accept the risk with a written reason. Submit updated source and every round is re-verified, so nothing closes unchecked.
Everything stays on record
Audits and the points they cost in one searchable timeline. Your source is never stored — we keep the record of the run, not your code.
Sign in with email, a connected wallet, or Google/GitHub — same account and history, whichever way you start.
Let your AI agent order its own audit
Give Claude, Cursor or your own agent an API key and it can price, run and collect a security audit by itself, over REST or as an MCP tool. Same audit and same report as the workspace, paid from your points.Give your AI agent a key and it prices, runs and collects audits itself, over REST or MCP. Same review as the workspace.
Create a key
In the workspace, open Agent API and create a key. Give it a daily spending cap, and revoke it the moment you no longer need it.
Plug it into your agent
Add grisoco.com/mcp to Claude, Cursor or any MCP client, or call the REST API from your own code.
It agrees a price first
Your agent asks for a quote and states the most it will pay, so it never spends more than it agreed to.
It acts on the findings
The audit runs in the background. Your agent collects the report as Markdown or JSON, and an audit that fails on our side is refunded automatically.
What should we fix, and what should we run?
Both boxes go to a person. You do not need an account to use either — but if you are signed in, we can reply.
Tell us what is working, and what is not
If a finding was wrong, misleading, or missing, that is the most useful thing you can send us.
Request a model
Is there an AI model you want behind your audits? Tell us which one — we count the requests, and the most requested are the ones we look at first.Tell us which AI model you want. The most requested get looked at first.
This one lives in the workspace. Requests are counted once per account, so we can tell how many people want a model rather than how many times a form was submitted — and asking again updates your note instead of adding a second vote.Requested from the workspace. One vote per account; asking again updates your note.
Request a model in the workspace