Kann Audits
View Audits Talk to an Expert

AI Security Analysis with Model Alpha

Kann AI / Model Alpha extracts Solidity call graphs and returns function-level security signals for expert validation. It is an early analysis tool, not an audit certificate.

Launch AI Scan Run Model Alpha Locally

How it works

From repository to security findings

Kann AI builds an understanding of your codebase, analyzes contract behavior and security-sensitive logic, and returns findings you can review and validate.

  1. 01
    Repository

    Load Repository

    Provide the smart contract codebase you want to analyze.

  2. 02
    Context

    Build Context

    Kann AI maps contracts, dependencies, inheritance, permissions, and protocol interactions.

  3. 03
    AI Analysis

    Analyze

    The model investigates code paths, assumptions, state changes, and security-sensitive behavior.

  4. 04
    Vulnerability Detection

    Detect Findings

    Potential vulnerabilities and suspicious behaviors are identified in context.

  5. 05
    Findings

    Review Output

    Receive structured findings with explanations engineers and security researchers can validate.

Local workflow

Use the official Model Alpha repository

Read the full README before installation and confirm code-handling requirements before analyzing confidential source.

Open installation guide Review source
Terminal
git clone https://github.com/Kann-Audits/model-alpha.git /tmp/model-alpha

Then follow the agent-specific dependency and installation instructions in the repository.

Hosted workflow

Run Kann AI directly through the website

Scan your repository through the Kann Audits interface without setting up Model Alpha locally.

KANN AIHOSTED SCAN
Repository
github.com/example/protocol
ModelModel Alpha
AnalysisFull repository
Run Security Scan
  1. 01Load Repository
  2. 02Build Context
  3. 03Analyze Contracts
  4. 04Detect Vulnerabilities
  5. 05Generate Findings

Trusted in production

What our clients have to say

Teams building real protocols use Kann AI as another layer in their security workflow.

“The scan was fast, easy to work with, and gave us useful security signals we could immediately investigate. We were impressed by what the model was able to identify.”

Fluton

“Kann AI did an excellent job on our codebase. We ran a full scan across 3,000+ lines of code, and it uncovered over 80 security findings. The results were impressive and gave us a lot of value throughout the audit scan.”

castr.fun

“Kann AI demonstrated how useful AI can become when it is built specifically for smart contract security. The output gave us meaningful additional context during our security review.”

Manifest Finance

Benchmarks

Model Alpha on DODO

See how Model Alpha performs against the DODO smart contract security benchmark.

Model Alpha Security Benchmark

DODO

DODO Cross-Chain DEX · 17 findings (5 High and 12 Medium)

Precision82.4%14 detected · 3 false positives
Detected vulnerabilities14 / 1782.4% detection coverage
High-severity coverage5 / 5100% detected
Missed findings3All Medium severity

Detection outcome

Detected14 · 82.4%Missed3 · 17.6%

Model Alpha detected every High-severity finding in the benchmark and nine of twelve Medium-severity findings.

Detection breakdown

SeverityExpectedDetectedMissedResult
High severity550Detected
Medium severity1293Partially detected
Source artifactDODO benchmark artifact · updated result supplied September 2, 2026

FAQ

AI Security Scan FAQ

Is Model Alpha a replacement for a manual audit?

No. It produces early function-level signals. Business logic, economic attacks, cross-contract behavior, and system assumptions still require expert review.

Which language does it support?

The current public Model Alpha repository documents Solidity contract analysis.

Is my code stored?

The public documentation says function context is sent to a hosted endpoint but does not publish a retention policy. Confirm handling requirements before submitting confidential code.

Can I run it locally?

The call-graph extraction and CLI run locally, while the documented analysis step calls a hosted Model Alpha endpoint.

How should a result be validated?

Inspect the complete call path, reproduce the behavior, test system assumptions, and have a qualified reviewer assess real impact before making a security decision.

Start with the scope

Let’s secure what you’re building

Share your scope, target date, and architecture. The security team will review the details and respond with the next steps.

Talk to an Expert View Audits