How AIGP Works
AIGP sits alongside your AI system. It keeps a record of important AI changes, checks them against rules, and explains its decisions. Your app and deployment tools stay under your control.
Platform Flow
AIGP connects the important parts of an AI change: what changed, what happened when it ran, how it was tested, which rules applied, and the final decision. It gives you one place to review that story.
- →
01
Your AI systems
Agents, applications, workflows, providers, deployment, and live traffic remain in your stack.
- →
02
Assets + evidence
Versioned managed assets and runtime observations establish governed facts and traceability.
- →
03
Evaluate + replay
Experiments, evaluation, comparison, and replay turn behavioral change into durable evidence.
- →
04
Govern + decide
Policy and dependency context produce explainable, versioned governance decisions.
- →
05
Analyze change
Enterprise change-assurance capabilities assess the consequences of a proposed AI change.
06
Assure + explain
Audit, lineage, events, and—where enabled—operational findings make the outcome reviewable and actionable.
This is the big picture. For details about the software pieces, see the Architecture Guide.
Three Governed Loops
01
Evaluation loop
Prompt / Model / Dataset → Experiment → Evaluation → Comparison → Recommendation
Which configuration behaves better?
02
Governance loop
Runtime evidence → Ontology → Policy → Decision → Evidence / Lineage
Was this AI activity compliant, explainable, and governed?
03
Change-assurance loop · Enterprise
Proposed change → Behavior Contract → Impact analysis → Replay / Evaluation → Release readiness
Is this change safe enough to proceed?
Extra enterprise features can build on these loops to show what needs attention and why.
One brand; explicit edition boundaries
AIGP uses the same governance model across its platform products. Evaluation, governed assets, policy, decisions, audit, replay, and lineage are OSS capabilities. Behavior Contracts, impact analysis, release readiness, and intelligence are Enterprise extensions and remain clearly identified as such.Cross-Cutting Control Planes
These workflows use the same shared services for access, records, and activity history. You can use them through Studio, REST, MCP, or events.
Ownership Boundary
Your stack owns
Application behavior, prompt deployment, model serving, workflow execution, infrastructure, CI/CD, and traffic control.
AIGP owns
Governance state, asset versions, evidence, policy versions, decisions, assurance records, audit context, events, and lineage.
A decision is not a deployment
Another release system can use an AIGP decision, but AIGP does not deploy a model or change your running application.Choose Your Next Step
- Run the 10-minute local tour to inspect seeded governed records.
- Follow the pre-release evaluation story from governed inputs through decision review.
- Read the Policy Engine guide to understand deterministic decisions.