AI Agent Governance

Govern an AI agent tool call.

A real AI agent requests an external action. RBEK decides whether it is authorized, executes the third-party API call through a governed boundary, and produces verifiable proof of what happened.

Try before the live example

Start with the zero-key governed execution proof.

The default developer proof requires no API key and no model inference. RBEK policy enforcement and RBEK evidence are real.

✓ Deterministic Execution Verified
curl -fsSL https://releases.rbekplatform.com/cli/stable/install.sh | bash && \
curl -fsSL https://raw.githubusercontent.com/rbekplatform/rbek/main/examples/real-governed-agent/demo.sh | bash

Zero-key mode: model inference not required · external governed action network execution: NO · policy enforcement: REAL · evidence: REAL.

Then continue below to the live example: real OpenAI model inference, a real Open-Meteo HTTPS action, receipt and certification.

Real execution

The agent asks. RBEK executes.

The AI model does not call the external provider directly. Execution authority remains inside the RBEK governed execution boundary.

01
OpenAI agent requests current weather
Real model inference using gpt-5-mini
PASS
02
RBEK creates the execution plan
Capability: weather.current
READY
03
Policy gate evaluates the action
Provider: open-meteo
AUTHORIZED
04
RBEK invokes the third-party HTTPS API
api.open-meteo.com
PASS
05
RBEK records execution evidence
Network and external API execution recorded
PASS
06
Receipt and certification are verified
Cryptographically bound execution proof
VALID
Agent: OpenAI / gpt-5-mini
Action: Get current weather
Third-party API: api.open-meteo.com
Execution mode: external-controlled

Policy authorization       PASS
HTTPS execution            PASS
Network execution          TRUE
External API execution     TRUE
Database execution         FALSE
Receipt                    VALID
Certification              VALID

Agent response:
Temperature: 25.9°C.
RBEK receipt valid: yes.
RBEK certification valid: yes.
Why RBEK

Governance stays between intelligence and execution.

Models can reason and request actions. RBEK controls whether and how those actions execute.

01 / Decide

Policy before execution

Actions pass through an explicit execution plan, dry-run and policy authorization gate before the provider can execute.

02 / Execute

Controlled provider boundary

The external call is performed by RBEK through the authorized provider boundary — not directly by the AI agent.

03 / Prove

Verifiable execution

RBEK records what executed and produces a verifiable receipt and promoted execution certification.

Provider-neutral

Weather is only the example.

Open-Meteo is used here because the execution is easy to verify publicly. The governed execution model is designed for real business systems.

CRM Customer actions
ERP Operational workflows
Payments Financial actions
Email Messaging workflows
Documents Document processing
Databases Controlled mutations
Internal APIs Enterprise systems
SaaS APIs External services
Govern execution

Let AI decide what it wants to do. Let RBEK decide what may execute.

RBEK is the governed execution layer for AI agents, workflows and software.