AI Governance
AI Gateway (AI Pilot managed from the Control Plane /pilot cockpit) provides enterprise AI governance across access, cost, safety, and data protection controls.
Recommended reading path
| Topic | Purpose | Link |
|---|---|---|
| AI Gateway | Routing, guardrails, MCP proxy from /pilot | AI Gateway |
| AI Agent Identity | NHI discovery, SPIFFE, /observe/nhi | AI Agent Identity |
| AI Observability | Langfuse traces via Bouncer OTEL | AI Observability |
| AI Gateway — full guide | Architecture and operator narrative | AI Gateway enterprise guide |
| NHI continuous verification | Agent Access Controls — discover NHIs from traffic, profile, enforce Rego (vendor-agnostic PIP) | NHI verification |
| MCP blast radius | Limit MCP tool exposure per agent | MCP blast radius |
| Agent framework connectors | Agentforce, Foundry, LangChain via one Bouncer | Framework connectors |
| Token cost circuit-breaking | BackendTrafficPolicy + Redis budgets | Token circuit-breaking |
| Capability matrix | CRDs vs Bouncer intercept vs PBAC | Capability matrix |
| Per-bouncer operations | Configure LLM/cost/safety by bouncer | AI Gateway management |
| Prompt control model | Enterprise prompt security controls and actions | Prompt Security Controls |
| Incident handling | Failure patterns and fixes | AI Gateway troubleshooting |
Governance model visual
Click to enlarge
Multi-framework agent governance
Control Core provides unified authorization for AI agents regardless of the orchestration framework. Whether agents are deployed through Salesforce Agentforce, Microsoft Azure AI Foundry, custom LangChain pipelines, or any other platform, the Bouncer intercepts and normalizes agent context into a single x-controlcore-agent-context header.
Click to enlarge
The Bouncer remains framework-agnostic — it forwards all vendor-specific headers into policy engine input generically. Controls authored in the Control Plane determine what each agent framework may access, which capabilities are permitted, and what data classification restrictions apply. This means you can enforce consistent governance across heterogeneous agent ecosystems without maintaining separate control sets per vendor.
What enterprises typically configure
- Model access restrictions by role/environment
- Token-aware budgets and fallback routing
- Prompt injection/jailbreak/exfiltration controls
- PII and secret redaction policies
- Exception handling with allowlists and trusted domains
- Agent capability restrictions per framework and identity
- Cross-framework data classification enforcement
- Temporal access windows for autonomous agents