RBEK

RBEK is a deterministic execution control layer for AI agents and software that need to perform governed real-world actions.

AI may propose an action. Deterministic software decides whether execution is authorized.

RBEK is designed around controlled execution, explicit authorization, policy-aware decisions, evidence, and provider-neutral integration.

Why RBEK

AI systems are increasingly able to decide what should happen next.

That does not mean they should automatically have authority to make it happen.

RBEK separates proposal from execution authority:

• AI or application logic may propose an action.
• Deterministic policy and authorization logic decide whether it may execute.
• Controlled execution produces evidence that can be inspected and audited.

Install

python -m pip install rbek

RBEK requires Python 3.11 or newer.

60-second quickstart

Verify the local installation:

rbek-cli doctor

Create a minimal local RBEK project:

rbek-cli init

Run the generated local workflow:

rbek-cli run

The onboarding path is:

install -> verify -> initialize -> run

These onboarding commands do not imply that an external provider or real-world side effect has been authorized.

First SDK example

from rbek.sdk import KernelBuilder

kernel = KernelBuilder().build()
capabilities = kernel.list_capabilities()

print(capabilities)

Expected output:

()

This first SDK path is provider-neutral and does not require network access, database access, or an external side effect.

CLI

RBEK's CLI exposes three practical levels.

Onboarding

rbek-cli doctor
rbek-cli init
rbek-cli run

Governance inspection

rbek-cli policy
rbek-cli workflow
rbek-cli authorize
rbek-cli adapter

Controlled execution

rbek-cli execution
rbek-cli evidence
rbek-cli provider

Use:

rbek-cli --help

for the complete command surface.

Telemetry is opt-in where supported.

Examples

Existing examples provide progressively more advanced integration paths:

• examples/refund-governance-demo — simple governed business-action example.
• examples/ai-agent-governed-tool-call — generic AI-agent controlled tool execution.
• examples/ai-agent-governed-external-tool — external-tool bridge pattern.
• examples/langgraph-governed-tool — LangGraph integration.
• examples/crewai-governed-tool — CrewAI integration.
• examples/public-v2-trusted-host — advanced trusted-host / API V2 integration.

Examples demonstrate integration patterns; they are not themselves normative execution authority.

Core execution model

RBEK follows a simple hierarchy:

1. deterministic software where possible;
2. explicit rules and known data;
3. AI or LLM reasoning where useful;
4. deterministic authorization before execution.

An AI-generated response is a proposal, not execution authority.

Links

• Homepage: https://rbekplatform.com/
• Repository: https://github.com/rbekplatform/rbek
• Documentation: https://rbekplatform.com/