← Back to all examples

CrewAI-style Multi-Agent Governance

CrewAI-style simulationPythonsimulationautomated tests

Researcher/Auditor/Purchaser roles calling Limetry evaluate over HTTP (no CrewAI SDK dependency).

Source folder: examples/crewai-multi-agent-financial-approvals

Documentation & Guide

README.md

crewAI-style financial approvals

Plain-Python Researcher / Auditor / Purchaser roles. The Auditor calls Limetry (PUT /v1/policies/{policyId} + POST /v1/policy/evaluate). The crewAI SDK is not installed. The workflow evaluates requests but does not purchase anything or create a human approval task.

Quickstart

cd examples/crewai-multi-agent-financial-approvals
python3 -m venv .venv
source .venv/bin/activate
pip install httpx pytest
pytest test_crew.py
python crew.py   # needs a running Limetry server

Policy highlights

  • Allow purchase for merchant_slack / merchant_github
  • Block merchant_malicious
  • max_cost_minor: $30.00

Scope and boundaries

DemonstratedBoundary
Multi-role simulationPlain Python role stand-ins
Real crewAI integrationNot included
Action governance HTTPReal calls in the example; mocked in tests
Financial approvalAllow/deny policy decisions only; no approval inbox
Purchase executionNot performed

Source Files

3 files in examples/crewai-multi-agent-financial-approvals
markdown
# crewAI-style financial approvals

Plain-Python Researcher / Auditor / Purchaser roles. The Auditor calls Limetry
(`PUT /v1/policies/{policyId}` + `POST /v1/policy/evaluate`). The crewAI SDK is
**not** installed. The workflow evaluates requests but does not purchase
anything or create a human approval task.

## Quickstart

```bash
cd examples/crewai-multi-agent-financial-approvals
python3 -m venv .venv
source .venv/bin/activate
pip install httpx pytest
pytest test_crew.py
python crew.py   # needs a running Limetry server
```

## Policy highlights

- Allow `purchase` for `merchant_slack` / `merchant_github`
- Block `merchant_malicious`
- `max_cost_minor`: $30.00

## Scope and boundaries

| Demonstrated | Boundary |
| --- | --- |
| Multi-role simulation | Plain Python role stand-ins |
| Real crewAI integration | Not included |
| Action governance HTTP | Real calls in the example; mocked in tests |
| Financial approval | Allow/deny policy decisions only; no approval inbox |
| Purchase execution | Not performed |