Ship AI agents you can defend in an audit
Every autonomous action verified, logged, and reversible — before it runs. The open-source Agent Control Plane with a 5-gate Verified Autonomy Pipeline. Policy-enforced. Tamper-evident. Self-hosted or managed.
Anthropic-compatible · EU AI Act-ready · MCP-native
Built on OpenAI · Anthropic · MCP · OpenTelemetry · FastAPI · Postgres
# pip install occp
from occp import ControlPlane
cp = ControlPlane(policy="policies/finance.rego")
result = cp.run(
agent="claude-opus-4-7",
task="Refund order #4521 if return window is open",
)
# PLAN → VERIFY → APPROVE → EXECUTE → AUDIT
print(result.audit_chain_id)- Plan
- Gate
- Execute
- Validate
- Ship
Works with your stack
- MCP
- OpenTelemetry
- Supabase
- Cloudflare
- GitHub
- Slack
The platform
Everything a governance-grade agent control plane needs
- Learn more →
Verified Autonomy Pipeline
Five deterministic gates — Plan, Gate, Execute, Validate, Ship — block any action that violates your policy before it reaches production.
- Learn more →
EU AI Act Art. 14 built-in
Human-in-the-loop escalation, transparent decision logging, and risk classification ship as first-class primitives, not bolt-ons.
- Learn more →
MCP-native tool catalog
Every tool in the catalog is typed, versioned, and schema-validated against the Model Context Protocol spec before agents can invoke it.
- Learn more →
Cost observability
Per-token USD spend tracked in real time via OpenTelemetry spans. Budget limits and alerts configurable per agent, per policy.
- Learn more →
Kill switch
Redis-backed global kill switch halts all running pipelines in under 200 ms. Circuit breaker per agent, per environment, per org.
- Learn more →
Audit chain
Append-only SHA-256 hash chain on every action event. Tamper detection on read. Exportable to S3 or any SIEM.
About us
Why a governance-first agent platform is built in Hungary
We started building OCCP in early 2026 with one conviction: small and medium businesses will only trust AI agents with real work once they can verify, log, and reverse every step. The first line of code was written on 21 February 2026, and ever since the codebase has been built around controllability — not the other way around.
Our mission
We put a platform in the hands of businesses where autonomous agents never act as a black box. Every decision is transparent, every action is traceable, and a manager can stop the process at any time. That is the governance-first approach: control first, speed second.
How we started
OCCP is the work of an engineering team that experienced the real obstacles to enterprise AI adoption — compliance, cost control, and distrust of automation — firsthand. We found the answer in a five-gate verification chain and a tamper-evident audit system.
Hungarian financial integration
The platform is built for the local business environment from day one, with connectors to NAV, Billingo, Wise, and MiniCRM so invoicing, accounting preparation, and customer handling can be automated within local regulation. Financial actions get a dedicated approval gate.
Governance first
Every feature is designed around verifiability and reversibility.
Local focus
Domestic financial systems, a native-language interface, and knowledge of local regulation.
Open core
The Community Edition is open source, so it stays transparent and verifiable by anyone.