mantık

Building in stealth

Technical audits of AI agents

Know what your agents actually do.

Examine their purpose, behaviour and controls. Trace each finding to evidence.

We start with technical audits in sensitive workflows. From that work, we’re building the platform for internal audit teams and audit partners.

We’re building Mantık in stealth. If you can help move it forward, there’s a trail to the people behind it.

Illustrative agent audit · 001

Supplier onboarding agent

Purpose & scope
Maintain supplier records. Review the model, source data, tools and permissions.
Evidence
Scenario tests, source checks and execution traces.
Open finding · F-01
A tool-call trace shows bank details changed without verified human approval.
Owner
Payments engineering
Retest
Pending a verified approval check at the tool boundary.
Scope → Evidence → Follow-upExample record
  1. Define the audit

    Establish the intended use, models, data, tools and permissions. Agree on the criteria to test.

  2. Test and trace

    Examine behaviour and output quality. Connect each finding to its requirement and technical evidence.

  3. Resolve and retest

    Assign owners to open findings. Verify fixes and retest when models, tools or permissions change.

mantık / field notes 001

Follow the evidence.

A short trail also lives in your browser console. Start with rabbit(), or follow it here.