Premiere Control Room¶
An agentic reliability engineer for live media premieres, built on the Google Agent Development Kit (ADK), Gemini, and the Grafana Cloud MCP server.
Live, unrepeatable media events — streaming premieres, award shows, live sports — break in ways that used to mean an engineer staring at a wall of Grafana dashboards under pressure. Premiere Control Room replaces that manual loop with a crew of five ADK agents — Sentinel, Detective, Producer, Responder, Wrap — that detect, correlate, brief, remediate, and document an incident automatically, with a forced human-approval gate before anything state-changing happens.
-
New here? Start with the Setup Guide
Get the backend and frontend running locally, then deploy to Google Cloud Run.
-
Already deployed? Read the User Guide
Signing in, roles, injecting incidents, approving remediations, reading the timeline.
-
How it's built
The agent crew, the FastAPI backend, the Next.js control room, and the design decisions behind them.
-
Security & governance
Role-based access control, the audit log, least-privilege Grafana MCP scoping, and secrets handling.
What's in the crew¶
| Agent | Role |
|---|---|
| Sentinel | Continuously monitors SLOs (rebuffer ratio, playback failure rate, origin error rate, encoder queue depth) and detects anomalies |
| Detective | Correlates signals across metrics, logs, and traces into a root-cause hypothesis, checking whether the same failure has happened before |
| Producer | Turns the finding into an executive-friendly incident brief and identifies who's on call |
| Responder | Proposes and executes remediation — high-risk actions always wait for human approval |
| Wrap | Generates the incident timeline and a downloadable postmortem report |
Source¶
The full source is on GitHub. Documentation source lives under docs/ in that repo — edits are welcome via pull request.