Scout, the AI Code Review Agent for All Complex Systems

Run static analysis, specification-guided fuzzing, static analysis for ZK circuits, and formal verification for ZK circuits in one click. Scout plans the review, validates findings with the right tools, and remembers your protocol between runs, so no review starts from zero.

// Every tool-assisted finding comes with the tool output behind it.
// Works on code that doesn't compile yet.

How the Scout AI code review agent works

Four steps in a loop. What you correct at the end is where Scout starts next time.

  1. // 001

    Scout learns your protocol

    Maps the architecture, builds a threat model and proposes the invariants to enforce.

    // BUILD THE MODEL
  2. // 002

    Scout hunts for exploitable bugs

    Looks for the bugs an attacker could use, within the credit budget you set.

    // TRACE THE PATH
  3. // 003

    Scout tests each finding

    Runs the right security tool on each one: Vanguard, OrCa, ZK Vanguard or Picus.

    // VERIFY WITH TOOLS
  4. // 004

    Scout remembers

    Keeps your corrections and the protocol model for the next run.

    // CARRY CONTEXT FORWARD

Corrections and findings are stored and used to improve the next run

Basic and Tool Assisted Code Reviewer

Start with a first read on any branch, then get tool-backed findings before an audit or a release.

Basic Code Review run: 11 findings (2 critical, 1 high, 3 medium, 4 low, 1 info) and the top three findings on ShieldedPool

Basic code reviewer

Reviews your code the way a human reviewer would, even on code that doesn't compile yet.

// BEST FOR EARLY BRANCHES
Tool Assisted Review: the Scout finding 'First depositor can inflate the share price' next to the OrCa counterexample, the exact transaction sequence that breaks the share price invariant

Tool assisted code reviewer

Plans and runs Vanguard, OrCa, ZK Vanguard and Picus against its threat model. Each finding comes with a location, severity, confidence score and the tool output, including counterexamples.

// BEST FOR AUDITS AND RELEASES

The AI Code Review Agent that learns your protocol

Tell Scout a violation is expected during asynchronous settlement, and it updates the invariant and the threat model, so the next run flags real bugs, not known behavior.

AI security review you stay in control of

Scout works within the credit budget you set and logs every task with why it ran, so you see exactly what it did with your code.

  • Set a credit budget

    Scout stops before going over and keeps every finding, even if the review didn't finish.

  • Read why each task ran

    Every tool Scout runs is logged with its reasoning.

  • Choose who sees it

    Findings and Scout's memory stay with the members of your AuditHub organization.

  • Remove a project

    Delete a project when you no longer need it. This action can't be undone.

Your code and Scout's memory of it never train a model.

Read our Trust page

Scout vs a general AI assistant

Scout runs on frontier models like other assistants, and adds the memory, the tools and the budget a chat session lacks.

Criteria Scout General AI assistant
Protocol understanding

Builds a threat model around your protocolMaps its architecture and proposes key invariants.

Works from the context provided in the conversation

Your feedback

Remembers your protocol between runs.Retains its threat model and your corrections for future reviews.

Feedback primarily guides the current conversation

Findings grounded in

Tests findings with specialized security tools.Uses static analysis, fuzzing, and ZK security tools where applicable, attaching evidence to findings.

Primarily model reasoning and provided context

Review budget

A credit budget you control.Set a budget for each review. Scout stops at the limit and keeps the findings already made.

Usage depends on the assistant and plan

Run artifacts: agent.log, findings.json, results/report.md and results/fio.json, with the results/fio.json viewer open

What your auditors receive

Scout hands your auditors the threat model, the invariants and the findings with their tool output, so the audit starts from a higher baseline. The audit still gives you a signed expert review.

AI code review agent FAQ

Everything you need to know about Scout, how it reviews your code, uses security tools, manages credits, and fits into your security workflow.

What is an AI code review agent?

An AI agent that reviews your code for bugs, the way a reviewer would. Scout goes further: it learns your protocol, tests its findings with security tools built on formal methods, and remembers what it learned.

Does my code need to compile?

No. In Basic Code Reviewer mode, Scout reasons about incomplete code. Tool-backed checks run on the parts that build, and you still get findings on the rest.

What can Scout review today?

Smart contracts and ZK circuits.

How are credits used?

Scout works within the credit budget you set and stops before going over it, keeping every finding it has made. Credits per plan are listed on the Pricing page.

Does Scout replace my security team?

No. Scout handles the groundwork so your audit starts from a higher baseline. Judgment on what matters stays with your security team and your auditors.

Still have questions? Contact us

Run your first review with Scout

Point Scout at your smart contracts or ZK circuits today.