The AI security auditor for onchain code
AI agents that run a complete audit of your smart contracts, then review every pull request with the full context of your protocol.
Sign-up is free, and new accounts get $50 in credit, good for up to half of an order.
Audits your protocol
It maps your contracts, roles and calls into a security graph and lists the invariants, the rules that must always hold.
Kept in your workspace
- invariants.mdwhat must hold
- recon.mdattack surface
- Findings5
- ReportsPDF, Markdown
Source code deleted after the audit
Mini DeFi Vault
Full audit with proof of concept
Completed
- Criticaldeposit() prices new shares on a balance that already includes msg.value, so depositors are under-minted and earlier holders take their principalPoC passed
- Criticalredeem() reduces totalShares by only shares/1e9, so remaining holders lose value and ETH is permanently lockedPoC passed
- CriticalBroken share accounting permanently freezes depositor principal and donations in MiniYieldVaultPoC passed
- Mediumredeem() checks the share count against minAssets instead of the ETH paid, so redemptions have no slippage protectionPoC passed
- MediumSearcher can sandwich a visible ETH donation and take yield meant for existing holdersPoC passed
- findings
- 5
- Critical
- 3
- proofs of concept passed
- 5 of 5
- start to finish
- 2h 7m
Reviews every pull request
Each pull request is checked against the security graph. The check fails on a High or Critical finding, or at the threshold you set.
lending-core
- #214 Adjust liquidation bonus1 High
- #213 Add pause guardianPassed
- #212 Bump oracle heartbeatPassed
Checked against the full audit of commit 3f9c2ab
QuillShield in numbers
90%
of the findings QuillAudits auditors reviewed were valid
Private beta, client Solidity code. Reported November 2026. What it covers
83.8%
of known vulnerabilities found on EVMbench, the OpenAI and Paradigm benchmark
Our own alpha run, July 2026. The best single model in the paper’s first release found 45.6%. How we ran it
350,000+
lines of Solidity reviewed across 155 projects
Private beta, until October 2026.
What users say
QuillShield beta user
I have run my repo through 2 other AI auditors to prepare for formal audits and to find out highs were missed is surprising. Glad I was able to test it out.
QuillAudits client
I’m very impressed by the workflow, PoC feature and layout of QuillShield. I can see how this will become an industry standard for smart contracts.
How QuillShield reviews every pull request
1
Audits your protocol
It maps your contracts, roles and calls into a security graph and lists the invariants, the rules that must always hold. Attach your whitepaper and past audits, and it checks the code against that intent.
Guide: run your first audit2
Attacks it like an adversary
Agents trace attack paths, and a reviewer agent tries to disprove each finding. On a full audit, each finding that holds up is tested with a proof of concept.
How findings are checked3
Reviews every pull request
Each pull request is checked against the security graph. The check fails on a High or Critical finding, or at the threshold you set.
Guide: check every pull request on GitHub
Built for code written by people and agents
GitHub App
A check and one comment, updated in place, on every pull request.
Pull request checksMCP for Cursor and Claude
Your coding agent reads findings and starts audits.
MCP setup$ npm install -g @quillaudits/cli
$ qs init
$ qs audit <repo> --watchCLI
Run and follow audits from the terminal.
CLI installGET /v1/audits/:job_id/findings
GET /v1/audits/:job_id/report
GET /v1/audits/:job_id/eventsREST API
Findings, reports and live events for any pipeline.
API referenceEscalate any finding to an auditor
Ask a QuillAudits auditor to validate it, or use the report as the starting point for a manual audit.
1,500+
protocols audited by QuillAudits since 2018
Clients include Taiko, StarkWare, ZetaChain, Metis and BrahmaFi.
Measured on EVMbench
EVMbench is a benchmark built by OpenAI and Paradigm from vulnerabilities found in real smart contract audits. In July 2026, QuillShield’s alpha found 98 of the 117 known vulnerabilities in it.
Each square is one known vulnerability. 98 found, 19 missed.
In 28 of the 40 projects, it found every known vulnerability.
Your unaudited code stays yours
Disclosed to you alone
Findings never go to a bug bounty platform or an audit contest.
Source code deleted after each audit
Reports, findings and the security graph stay in your workspace until you delete them.
Never used for training
No model, ours or a third party’s, is trained on your code or your findings.
Read-only access to your code
The GitHub App reads your code and writes only checks, statuses and pull request comments.
ISO/IEC 27001:2022
QuillAudits as a company runs a certified information security management system, audited every year.
Latest from the team
7 minute read
11 valid bugs in 20 live bounties and audit contests
11 findings judged valid, 4 filed as High, in 20 live Solidity bug bounties and audit contests from March to November 2026. 18 more are in judging.
3 minute read
QuillShield found a bug an independent researcher had ruled out
He had ruled a bug out after a safeguard stopped his attack. QuillShield found a path past the safeguard, and he reproduced it on a mainnet fork.
6 minute read
QuillShield found 98 of 117 known vulnerabilities on EVMbench
The QuillShield alpha found 98 of 117 known vulnerabilities on EVMbench (83.8%) in our own run on 8 July 2026. What EVMbench measures and how we ran it.
Product update
Claude Mythos 5.2, Workflow Studio and your own AI
QuillShield now supports Claude Mythos 5.2.
Questions teams ask first
How do I start, and when do I pay?
Sign up for free, import a repo and choose what to audit. You see the exact price before anything is charged, and the audit starts when you approve it. The $50 credit for new accounts covers up to half of an order and expires after six months.
Guide: run your first auditDoes QuillShield replace a manual audit?
No. QuillShield is an AI smart contract audit that runs between manual audits. It reviews every change, and you can request a manual audit from QuillAudits.
Which languages does QuillShield support?
Solidity is supported today, for Foundry projects. More languages are coming, one at a time.
How long does a QuillShield audit take?
The demo vault at the top of this page took 2h 7m, and an independent researcher’s audit of a live lending contract took 2h 30m. Larger codebases take longer. Findings appear while the audit runs, so your team can start fixing before it finishes.
How are findings checked before they reach me?
A separate reviewer agent tries to disprove every finding before you see it. Each one that holds up arrives with the exact lines, the impact, a recommended fix and, on full audits, the proof of concept result. If you want a human check as well, you can ask a QuillAudits auditor to verify specific findings from the audit page.
How expert review worksCan a failing check stop a merge?
Yes, on GitHub, where your plan includes branch protection or rulesets. Make the QuillShield check required and a pull request cannot merge while it has a High or Critical finding, or one at the threshold you set. Each full audit comes with five pull request checks, and more can be added. They need a GitHub or GitLab.com repo, so an uploaded zip gets a full audit and no checks.
QuillShield is building the security layer between AI and money, so you can keep shipping between audits.
Sign-up is free, and new accounts get $50 in credit, good for up to half of an order.
- 1Sign up and import a repo
- 2See the exact price before anything is charged
- 3Approve it, and the audit starts