Skip to content
GOVERN Probe

Quickstart

Prerequisites

  • Docker installed and running
  • A GOVERN API key (get one at govern.archetypal.ai)
  • An application making AI API calls

Step 1 — Pull and run

Terminal window
docker run -d \
  --name govern-probe \
  -p 4020:4020 \
  -e GOVERN_API_KEY=gvn_live_xxxxxxxxxxxx \
  -e GOVERN_ORG_ID=org_xxxxxxxxxxxx \
  -e UPSTREAM_URL=https://api.anthropic.com \
  archetypal/govern-probe:latest

The Probe is now running on port 4020, forwarding to Anthropic.

Step 2 — Redirect your app

Change one environment variable in your application:

Terminal window
# Before
ANTHROPIC_BASE_URL=https://api.anthropic.com


# After
ANTHROPIC_BASE_URL=http://localhost:4020

Your app code stays identical. The SDK uses the base URL automatically.

Step 3 — Make an inference

Run your application normally. The Probe intercepts the inference and begins scoring.

import anthropic


client = anthropic.Anthropic(
    base_url="http://localhost:4020",  # <- Probe endpoint
    api_key="your-anthropic-key",
)


message = client.messages.create(
    model="claude-sonnet-4-20250514",
    max_tokens=1024,
    messages=[{"role": "user", "content": "Summarize our Q3 results."}]
)
print(message.content)

Step 4 — Verify it’s working

Check the Probe health endpoint:

Terminal window
curl http://localhost:4020/healthz
# {"status":"ok","version":"1.0.0","upstreamReachable":true,"telemetryConnected":true}

Check the metrics endpoint:

Terminal window
curl http://localhost:4020/metrics
# govern_inferences_total 1
# govern_inferences_scored_total 1
# govern_telemetry_batches_flushed_total 1
# govern_proxy_latency_p50_ms 3.2

Step 5 — Open the GOVERN dashboard

Navigate to your organization in govern.archetypal.ai. Within 10 seconds of your first inference, you will see:

  • The inference logged with full metadata
  • Security, bias, accuracy, drift, and cost scores
  • Latency and token usage metrics

What just happened

Your App → POST /v1/messages → GOVERN Probe → POST /v1/messages → Anthropic
                                    │
                                    ├── forwarded response immediately (p50: 3ms overhead)
                                    │
                                    └── scored asynchronously → buffered → flushed to GOVERN

Common provider configurations

OpenAI

Terminal window
docker run -d \
  -p 4020:4020 \
  -e GOVERN_API_KEY=gvn_live_xxxx \
  -e GOVERN_ORG_ID=org_xxxx \
  -e UPSTREAM_URL=https://api.openai.com \
  archetypal/govern-probe:latest
from openai import OpenAI
client = OpenAI(base_url="http://localhost:4020/v1", api_key="your-openai-key")

Ollama (local models)

Terminal window
docker run -d \
  -p 4020:4020 \
  -e GOVERN_API_KEY=gvn_live_xxxx \
  -e GOVERN_ORG_ID=org_xxxx \
  -e UPSTREAM_URL=http://host.docker.internal:11434 \
  archetypal/govern-probe:latest

Next steps