ContinuityOS

Autonomous production intelligence for film crews.

ContinuityOS is an AI production supervisor for film and media sets. It watches operational telemetry, investigates continuity and technical anomalies, and gives the crew an evidence-backed go / no-go decision before the next shot.

Our judge demo uses a fictional production called Project Eclipse and follows one clear workflow:

AT RISK → investigate → HOLD SHOOT → resolve → READY TO SHOOT

The problem

Film sets are highly coordinated environments where small mistakes can become expensive. A continuity mismatch, camera problem, or schedule overrun may be visible in different systems and to different crew members, but the decision to keep shooting still has to happen quickly.

ContinuityOS turns those scattered signals into one production decision.

What it does

During Scene 27, ContinuityOS detects three production risks:

• a wardrobe continuity mismatch
• Camera B dropped frames above threshold
• the scene running 23 minutes behind schedule

When the director asks “Can we safely move to Scene 28?”, the production supervisor investigates the available evidence, checks runtime telemetry, compares continuity state, evaluates operational risk, and returns a clear recommendation.

For the demo scenario, the answer is HOLD SHOOT until the issues are resolved. Once corrected, the system moves to READY TO SHOOT.

How we built it

The runtime architecture is:

Production UI → Google ADK / Gemini supervisor → Grafana Cloud MCP → logs, metrics, alerts → production decision

Gemini

Gemini performs the reasoning step across continuity, camera, and schedule evidence and produces the final structured production recommendation.

Google ADK

Google Agent Development Kit defines the production-supervisor agent and orchestration contract used by the runtime.

Grafana Cloud MCP

ContinuityOS calls the Grafana Cloud MCP endpoint at runtime so the agent can access monitoring tools and observability context rather than treating Grafana as a decorative dashboard.

Safe demo fallback

The application also contains a deterministic Project Eclipse fallback so the public demo remains understandable if credentials are unavailable. The fallback is explicitly labeled and never claims that a live Gemini or Grafana request occurred. With credentials configured, the API reports real_runtime and the Grafana MCP connection state.

Why it matters

Most AI film tools focus on generating content. ContinuityOS focuses on the operational layer of production: helping crews detect problems before they become delays, unusable footage, or reshoots.

The broader idea is an AI incident commander for physical media production—combining observability, reasoning, human approval, and production context in one workflow.

What we learned

The key design lesson was that an agent is more useful when it has a bounded operational decision to make. Instead of building a generic film chatbot, we designed one high-value loop: inspect the evidence, explain the risks, and recommend whether production should proceed.

What's next

Future versions could connect continuity metadata, camera telemetry, call sheets, shot lists, equipment health, production schedules, and approved crew workflows so the supervisor can support an entire shoot day while keeping humans responsible for consequential decisions.

Built With

• css
• gemini
• google-adk
• google-cloud
• grafana-cloud-mcp
• html
• httpx
• javascript
• python
• vercel