Since we launched Remix, the mission has been to make building, deploying, and testing smart contracts as seamless as possible, and part of that has always meant helping you ship contracts that are free of vulnerabilities.
First, with Remix Analysis and Solhint for linting and basic checks, then Slither for comprehensive static analysis. Now we’re adding AI audits.
Your AI agent (or you) can now run a complete preliminary security review of a Solidity contract through Remix. No Remix account, API key, or human in the loop. It takes two calls and starts at twenty cents.
This is a preliminary review that catches common classes of bugs fast and cheaply. You still need a professional audit before your contract hold real value.
The problem with regular LLM checks
You have probably pasted a contract into an LLM and asked it to find the bugs. It works, up to a point. You’ll get back a clear, well formatted list of findings, some of them real, some not, with no reliable way to tell which is which.
The LLM model has your code, but nothing is telling it what to look for. Asked to find bugs, it falls back on what it has already learned: reentrancy, overflow, access control. Those checks are useful, but they’re generic and skew toward older Solidity.
Researchers at UNSW Sydney and CSIRO's Data61 found recall on some vulnerabilities fell to 13% on Solidity 0.8 compared to 0.4, mainly because the models leaned on spotting newer libraries and frameworks. The same study found a better-framed prompt could cut false positives by over 60%. Your vault has invariants a staking contract does not, and a generic sweep is looking for neither.
What helps is a better question. Give the same model a checklist scoped to your contract, and it works through the bugs that apply instead of the ones it read about. Remix runs this as two endpoints concerning Security Checklists.
Auditing with Remix
Two calls, both taking your raw Solidity sources and the model you want to run them on.
The model sets the price. Pick one per request with the model field, which is required and is case-sensitive:
Get the checklist strips your contract to a skeleton (with only the function declarations and without the function bodies), and maps that skeleton onto a security checklist taxonomy.
Audit takes your sources and the checklist and works through it. Findings come back classified CRITICAL through INFORMATIONAL, each with a location, an impact, and a remediation step, plus an executive summary.
The checklist step is what separates this from pasting a contract into an LLM. With the checklist, the model gets a list of the things worth investigating on your contract.
Paying for the calls
Each call is paid inline over x402, an open standard that uses the HTTP 402 status code. Your agent calls the endpoint and gets a 402 back with the terms:
{
"x402Version": 2,
"accepts": [
{
"scheme": "exact",
"network": "eip155:8453",
"amount": "150000",
"asset": "0x833589fCD6eDb6E08f4c7C32D4f71b54bdA02913",
"payTo": "0x4CF33cCB14B43D76b4f672C3feDe297d1d763ace",
"maxTimeoutSeconds": 300
}
]
}That’s a DeepSeek audit. eip155:8453 is Base, the asset is the USDC contract there, and the amount is in USDC’s six decimals, so 150000 is $0.15.
Your agent signs an authorization for that amount and sends the request again. A facilitator settles the payment on chain and covers the gas, so you pay the quoted price and nothing more.
Each step takes a minute or two, so the paid request returns a job ID straight away and your agent polls /job/<id> for the result. Polling is free. A finished job can only be read once, and an unread result is discarded after five minutes, so save it as soon as it arrives.
Quickstart
To run an audit with an agent, you need Coinbase’s Agentic Wallet, Node.js version 24 or later, and an AI agent that can run terminal commands, such as Claude Code.
Install the agentic wallet by running the command below:
npx awal status
# optionally add the skills
npx skills add coinbase/agentic-wallet-skills
After your installation, run the command below to open and set up the wallet:
npx awal showFollow the instructions, create an account, and sign in with an email instead of a seed phrase. The private keys stay in Coinbase’s infrastructure, so your agent can spend from the wallet without ever seeing a key.
Next, add funds from the same window. Use the Fund button to buy USDC through Coinbase Onramp where it’s available, or send USDC on Base to the address that npx awal address prints. Five dollars covers a dozen Sonnet audits.
Then give your agent the prompt below or similar:
Run npx awal status to check you can access my wallet. Then audit
<PATH_TO_CONTRACT> with Remix's x402 endpoints at
https://api.remix.live/mcp/x402-http and use model "Sonnet". Check each
endpoint's schema with npx awal x402 details first.
Call get_audit_checklist, then pass its markdown to do_audit. If a call
returns a job ID, poll /job/<id> every 5 seconds until it's done, and save
the result right away because it can only be read once. Report the
findings by severity.
The idea behind the prompt is that your agent first checks it can access your awal wallet (npx awal status). You then point it to Remix’s x402 endpoints (https://api.remix.live/mcp/x402-http) and choose a model (use model "Sonnet"), which can be any of the three in the table above. Last is the flow. The agent gets the checklist, passes its markdown to the audit, and polls each job until the report is ready (Call get_audit_checklist, then pass its markdown to do_audit).
The schema check (npx awal x402 details) shows what each endpoint expects before the agent pays. A request missing a required field is rejected before any payment settles, but a contract that doesn’t compile still costs the full fee, so make sure the contract compiles!
Applications and Suggested Workflows
Here are a few ways to use the audit.
Auditing while you build
Ask your agent to audit what it just wrote. It can read the report, fix the findings, and run again in the same session. Use DeepSeek for these passes. At 20 cents a run, you can check each significant change.
You don’t need a new checklist every time, either. /do_audit accepts any checklist in Markdown, so if the contract’s shape hasn’t changed, reuse the one you have and pay only for the audit. The generated checklist comes from your contract’s declarations, so it doesn’t know your intended behavior. You can add your protocol’s invariants to it, or write your own.
Pairing the audit with Slither
Slither and the AI audit catch different classes of problems, and Remix serves both from the same place. This prompt runs them together:
Use Remix's x402 endpoints at https://api.remix.live/mcp/x402-http.
1. Compile <PATH_TO_CONTRACT> with /compile and stop if it fails.
2. Run /analyze to get Slither's findings.
3. Run get_audit_checklist and do_audit with model "DeepSeek".
4. Merge the results, remove duplicates, and rank the findings by severity.
The whole run costs 23 cents.
Preparing for a professional audit
Run Fable before your code freeze and fix the common bugs, so your auditors can spend their time on your protocol’s logic.
The two steps don’t have to use the same model. A DeepSeek checklist with a Fable audit costs 75 cents.
Running audits in CI
The endpoints are plain HTTP, so you can call them from your own code with @x402/fetch. A pipeline can run an audit on each pull request that touches your contracts and post the report as a comment.
Use the report as review input rather than a pass/fail check, because model output can vary between runs. Pay from a dedicated wallet with a small balance, and keep its private key in your CI secrets.
The AI audit documentation has the full parameters, response formats, and a working code example.
Where this fits
Solhint and Slither catch the patterns they know, and plain AI audits catch whatever the model happens to look for. Using these endpoints gives the model a checklist built from your contract. And because your agent pays for each call over x402, the whole audit runs without a human in the loop, from twenty cents a run. It won’t replace a professional audit, but it will catch the easy bugs before your auditors have to. Try it on your next contract and tell us what it finds.

