Tutorial: Five-minute gateway¶
In five minutes you will have a local Aegis gateway running, make a governed chat request through it, and inspect the audit log.
Prerequisites¶
- Python 3.11+
- uv (recommended) or pip
1. Install¶
pip install aegis-gateway
2. Initialise¶
aegis init # writes aegis.yaml in the current directory
The generated aegis.yaml enables PII masking and leaves everything else as
commented examples. For this tutorial, replace its contents with the minimal
working config below:
providers:
demo:
type: openai_compatible
base_url: http://localhost:8000/v1
api_key: demo
routes:
default:
provider: demo
3. Start the dev server¶
aegis dev # binds localhost:8000, no auth, FakeProvider
The dev command always uses FakeProvider — a safe in-memory provider that
returns a canned response without making any real model calls. Leave this
terminal running.
4. Send a governed chat request¶
In a second terminal:
curl -s http://localhost:8000/v1/chat/completions \
-H "Content-Type: application/json" \
-d '{"model":"default","messages":[{"role":"user","content":"Hello!"}]}' \
| python3 -m json.tool
You will see an OpenAI-compatible response. Every request passes through the Aegis pipeline — ingress guards, route resolution, execution, egress guards — even with FakeProvider.
5. Check the audit log¶
curl -s http://localhost:8000/v1/audit | python3 -m json.tool
The runs array contains one entry for every request, with status, principal,
route, and verdict events.
Next steps¶
- Write your first guardrail — add a custom policy
- HITL approvals — pause requests for human review
- Deployment guide — production setup