continuity-engine

Human-governed continuity intelligence and production memory for creative production workflows.

STUDIO//ONE Web Console

STUDIO//ONE now includes a minimal judge-facing FastAPI console for the canonical workflow:

BRAINSTORM -> REFINE -> FINALIZE STORYBOARD -> GENERATE ASSETS -> QUALITY CONTROL -> POST PRODUCTION -> PUBLISH

Local Setup

Create a virtual environment, install dependencies, and configure local environment values from the placeholder template:

python -m venv .venv
.\.venv\Scripts\python.exe -m pip install -r requirements.txt
Copy-Item .env.example .env

Fill .env with reviewer-owned Google Cloud, Gemini, ClickHouse, and Secret Manager values. Do not commit .env or credential files.

Required runtime environment variables:

• GOOGLE_CLOUD_PROJECT
• GOOGLE_CLOUD_LOCATION
• GEMINI_MODEL
• CLICKHOUSE_HOST
• CLICKHOUSE_USER
• CLICKHOUSE_DATABASE
• CLICKHOUSE_PASSWORD_SECRET

Optional runtime environment variables:

• CLICKHOUSE_PORT defaults to 8443
• CLICKHOUSE_SECURE defaults to true
• CLICKHOUSE_VERIFY defaults to true
• CLICKHOUSE_MCP_QUERY_TIMEOUT defaults to 90
• MCP_CLICKHOUSE_PYTHON overrides the Python executable used for the MCP subprocess

Local Web Run

.\.venv\Scripts\python.exe -m uvicorn studio_one.web.app:app --host 127.0.0.1 --port 8000

Then open http://127.0.0.1:8000.

Health endpoints:

• /health verifies the FastAPI process and canonical stage count only.
• /ready verifies required environment variable presence and installed mcp-clickhouse availability. It does not call Gemini, Secret Manager, or ClickHouse.

The web console does not generate media, edit final video, authenticate to external platforms, upload media, schedule posts, or publish content. PUBLISH prepares a creator-ready package for approval and manual posting only.

Docker Build and Local Container Run

docker build -t studio-one-cloud-run .
docker run --rm -p 8080:8080 -e PORT=8080 studio-one-cloud-run

Open http://127.0.0.1:8080/health for a health-only container check. Full workflow calls require runtime environment variables and Google Cloud identity.

Google Cloud Run Deployment

The FastAPI production entrypoint is studio_one.web.app:app. The container starts Uvicorn on 0.0.0.0 and uses Cloud Run's $PORT.

Build and push with Cloud Build:

gcloud builds submit --tag REGION-docker.pkg.dev/PROJECT_ID/ARTIFACT_REPOSITORY/studio-one:TAG .

Deploy to Cloud Run:

gcloud run deploy studio-one `
  --image REGION-docker.pkg.dev/PROJECT_ID/ARTIFACT_REPOSITORY/studio-one:TAG `
  --region REGION `
  --service-account SERVICE_ACCOUNT_EMAIL `
  --allow-unauthenticated `
  --set-env-vars GOOGLE_CLOUD_PROJECT=PROJECT_ID,GOOGLE_CLOUD_LOCATION=us,GEMINI_MODEL=gemini-3.5-flash,GOOGLE_GENAI_USE_VERTEXAI=true,CLICKHOUSE_HOST=CLICKHOUSE_HOST,CLICKHOUSE_USER=CLICKHOUSE_USER,CLICKHOUSE_DATABASE=continuity_engine,CLICKHOUSE_PASSWORD_SECRET=CLICKHOUSE_PASSWORD_SECRET,CLICKHOUSE_PORT=8443,CLICKHOUSE_SECURE=true,CLICKHOUSE_VERIFY=true,CLICKHOUSE_MCP_QUERY_TIMEOUT=90

Minimum runtime IAM for SERVICE_ACCOUNT_EMAIL:

• Secret Manager access to the configured ClickHouse password secret: roles/secretmanager.secretAccessor on that secret.
• Vertex AI/Gemini invocation in the configured project/location. Prefer a custom role limited to prediction/inference permissions where available; otherwise use the narrowest predefined Vertex AI/Agent Platform user role accepted by the project.

Cloud Run should use its service account and Application Default Credentials. Do not deploy with downloaded service-account JSON keys in the image or repository.

Tests

.\.venv\Scripts\python.exe -m unittest discover -s tests -v