How it works
AIGRCeler runs inside Claude Code as a team of agents and skills. It leads each stage, writes to the people involved in plain business words, and keeps a tamper-evident record of what your company decided about each AI system.
The whole architecture in 8 minutes.
Why small and medium companies struggle with AI governance, risk and compliance work, and how AIGRCeler speeds it up while people keep every decision.
More films on our YouTube channel. The narration is AI-generated.
Four stages for each AI system
- Stage 1
Discover
What the system is, what it does, who it affects, what data it uses.
- Stage 2
Assess
How risky it is, and which controls it needs.
- Stage 3
Govern
The controls are approved by AI Governance and given owners.
- Stage 4
Operate
Owners send proof, the proof is checked, risk is re-scored, incidents are handled.
Operate never closes, because governing a live system never ends.
Who does what
The agents draft, ask, check, score and record. People decide. AIGRCeler never approves a control, accepts a risk or starts a stage on its own. AI Governance is the name used here for the people in your company who are responsible for AI.
What you get
- 11 agents and 11 skills for Claude Code that lead each stage.
- An engine written in Python's standard library only. It writes every record, so an agent cannot hand-edit a status.
- A starter catalog of 104 controls in seven areas, 32 risk scenarios and a 77-question discovery profile. The company owns its copy and changes it through proposals.
- Evidence that is fingerprinted when it arrives and expires on a date or when the model version changes. Only verified controls earn credit against risk.
- A decision journal that cannot be changed without detection: one line per decision, in the decision-maker's own words.
- A governance dashboard built by the engine from a fixed template.
- A map of where the controls meet ISO/IEC 42001, the EU AI Act and the NIST AI RMF 1.0. It shows where the controls help and never says that a company complies.
- A self-test of the engine that you can run yourself.
Several settings (retention periods, incident clocks, risk bands, evidence validity, legal mappings) are starting values. They must be checked for your company before you rely on them.