AetherBrowse Cloud Run Quickstart

This moves your local browser agent (agents/browser/main.py) to Cloud Run so your AI agents can call it remotely.

1) Deploy browser service

chmod +x deploy/gcloud/deploy_aetherbrowse.sh
./deploy/gcloud/deploy_aetherbrowse.sh <GCP_PROJECT_ID> us-central1

Outputs:

• Service URL
• Health URL (/health)
• Browse URL (/v1/integrations/n8n/browse)

2) Get API key from Secret Manager

gcloud secrets versions access latest \
  --secret=scbe-browser-api-key \
  --project=<GCP_PROJECT_ID>

3) Run 2-5 agent jobs in parallel

export SCBE_API_KEY="<copied-secret-value>"
python scripts/aetherbrowse_swarm_runner.py \
  --jobs-file examples/aetherbrowse_tasks.sample.json \
  --url "https://<cloud-run-host>/v1/integrations/n8n/browse" \
  --concurrency 3 \
  --save-screenshots-dir artifacts/screenshots \
  --output-json artifacts/swarm_run.json

The runner now emits DecisionRecords + traces per job:

• artifacts/aetherbrowse_runs/<run_id>/decision_records/*.json
• artifacts/aetherbrowse_runs/<run_id>/traces/*.json

4) Use from n8n

• Node: HTTP Request
• Method: POST
• URL: https://<cloud-run-host>/v1/integrations/n8n/browse
• Header:
  • X-API-Key: <scbe-browser-api-key>
  • Content-Type: application/json
• Body:

{
  "workflow_id": "wf-001",
  "run_id": "run-001",
  "source": "n8n",
  "actions": [
    {"action":"navigate","target":"https://example.com","timeout_ms":12000},
    {"action":"screenshot","target":"full_page","timeout_ms":12000}
  ],
  "session_id": "agent-1"
}

Operational Notes

• This gives you remote browser execution; your local machine can stay small/off.
• Keep concurrency at 2-5 first to control cost and stability.
• Cloud Run may cold-start; first call can be slower.
• Rotate scbe-browser-api-key regularly.