Enterprise Agent Control Plane
A centralized governance plane that registers, authorizes, rate-limits and audits every AI agent call before it reaches a protected tool.
Built by naman · AI Developer Intern
The system, up close
Real captures from the running system — the demo, screenshots and architecture come first, then the deep dive.
Screenshots
Architecture
Why this needed to be built
AI agents are being handed credentials and tools, but most systems have no centralized enforcement — an agent that drifts beyond its brief can call tools and read data it was never meant to touch.
Without a single control point there is no consistent authorization, no kill switch and no audit trail, which makes agents a compliance and security liability rather than a lever.
What we built, and how it changes the workflow
We built a control plane that sits in front of every agent action: identify the agent, check its status, run the policy engine, and only then allow or deny the tool call — logging everything.
Every request flows through a gateway: agent identification → status check → policy engine → tool permission → data scope → rate limit → allow/deny → audit log. The agent never talks to a protected tool directly.
A dashboard exposes the registry, policies, audit trail and a kill switch, while RBAC governs who can change what. Mock tools demonstrate the enforcement so the security properties can be validated without live infrastructure.
Core design principle
- Every tool call passes policy with no bypass path
- Audit logging is immutable and attached to every decision
- RBAC is enforced server-side on every endpoint
What it can do
A focused set of capabilities that define the system — no marketable filler, only what actually runs.
Agent Registry
Central catalog of every registered agent and its declared capabilities.
Policy Engine
Declarative rules that decide allow/deny per agent, tool and data scope.
Control Gateway
Single entry point every agent call must pass through.
Tool Authorization & Data Scope
Per-tool permission and per-dataset access checks.
Rate Limiting
Bounds how often an agent can act (in-memory in the MVP).
Kill Switch & Audit Trail
Instantly disable an agent and record every decision for compliance.
RBAC Dashboard
Manage agents, policies and reviews with role-based access.
The end-to-end workflow
The system processing path, in order.
- 1
Agent identifies
The gateway resolves the calling agent from its credential.
- 2
Status check
Disabled or unknown agents are rejected before any policy runs.
- 3
Policy evaluation
Tool permission, data scope and rate limit are checked in order.
- 4
Allow / Deny
Only requests passing all checks reach the protected tool.
- 5
Audit log
Every decision is recorded with agent, action and outcome.
- 6
Reviewer oversight
RBAC users manage agents and policies from the dashboard; a kill switch can disable any agent.
Stack & tech choices
- Backend
- Python, FastAPI, SQLAlchemy, Pydantic, PostgreSQL, Alembic, PyJWT
- Frontend
- React, TypeScript, Tailwind CSS
- Infra
- Docker, Docker Compose
- Testing
- Pytest
Tested, verified and honestly scoped
Validation & tests
Validated against a security matrix of agent/policy/tool combinations; see docs/validation.md for the full scenario set.
- Gateway rejects unknown/disabled agents before policy runs
- Tool + data-scope + rate-limit checks run in strict order
- Kill switch disables an agent mid-session
Security & hardening
- JWT-based agent identity
- Server-side RBAC on every endpoint
- All decisions written to an append-only audit trail
Honest limitations
Enterprise Agent Control Plane is an internally developed engineering prototype. We state limitations rather than overclaim production readiness.
- In-memory rate limiting in the MVP
- Mock tools for demonstration only
- No multi-tenancy or enterprise SSO yet
Deep-dive references
Documentation
API, architecture, security and validation references.
Where this can go next
Realistic next steps we've scoped — not aspirational marketing copy.
- Redis-backed rate limiting
- Cryptographic audit chaining
- Multi-tenancy
- Enterprise SSO
- Anomaly detection
- Cost tracking
Have a similar workflow?
Zenera Labs can build a custom version for your business.
This is one of naman's internally developed systems. Tell us about your requirement and we'll scope a tailored build around it.
Other systems we've built internally
AI Contract Analysis & Risk Detection
Faster document review with traceable evidence — upload a contract and get risk findings with source-verified citations.
naman · Engineering Portfolio
View case study
SentinelAI — Security Log Anomaly Detection
Risk-scored alerts from security events — a full-stack Mini-SIEM with ML detection, attack-chain correlation and explainable alerts.
pragna · Engineering Portfolio
View case study
AI Data Analyst & Business Intelligence Agent
Evidence-grounded business intelligence from spreadsheets — upload CSV/XLSX, get analysis, visualizations and explained answers.
priyanka · Engineering Portfolio
View case study