Install with Docker Compose#

Use the v1.1.0 NGC release bundle to run POC Factory from published images on a trusted single host. The bundle includes the Compose file, environment template, and launch preflight. Python, Node.js, and a source checkout are not required to run the released application.

Prerequisites#

Prepare the host, release package, and credentials before starting the stack.

  1. Install Docker with the Compose plugin and reserve at least 8 GB RAM, 20 GB free disk, and host ports 80 and 8000.

  2. Download the v1.1.0 Compose deployment bundle and image pull access for your NGC organization.

  3. Obtain an OpenAI-compatible inference API key, or plan to save a complete user inference profile in Settings.

  4. Plan where to retain the encryption key, PostgreSQL password, generated POCs, and backups.

The Compose stack mounts /var/run/docker.sock to validate generated candidate containers. Run it only on a trusted host with access to that daemon.

Installation Methods#

Use the MCP profile for the application stack, then add Phoenix when tracing is needed.

MCP profile#

The mcp profile starts the backend, frontend, PostgreSQL, and GitHub MCP proxy. The proxy can improve source and blueprint lookups when a GitHub token is configured.

  1. Extract the v1.1.0 NGC resource package and open its poc_factory/ directory. Confirm README.md, env.example, docker-compose.yml, and scripts/validate-compose-env.sh are present.

  2. Copy the environment template:

    cp env.example .env

  3. Edit .env for the release and your deployment:

    NIM_NGC_ORG=nvidia
VERSION=1.1.0
INFERENCE_API_KEY=<your-inference-key>
ENCRYPTION_KEY=<stable-fernet-key>
POSTGRES_USER=<service-user>
POSTGRES_PASSWORD=<strong-password>
POSTGRES_DB=pocfactory_db
ENVIRONMENT=development
NGC_AUTH_ENABLED=false
SECURE_COOKIES=false

    Preserve the shipped model role values unless you have verified replacements. Generate a Fernet-compatible key with openssl rand -base64 32 | tr '+/' '-_' | tr -d '\n' and store it securely. Changing the key later makes saved credentials unreadable.

  4. If image pulls require authentication, run docker login nvcr.io --username '$oauthtoken' and enter the NGC personal API key at the password prompt.

  5. Inspect and start the pinned images:

    docker compose --profile mcp config --images
docker compose --profile mcp pull
docker compose --profile mcp up -d
docker compose --profile mcp ps

The image listing should show poc-factory-mcp, poc-factory-frontend, and github-mcp-proxy-nat-agentic-poc from your NGC organization, all tagged 1.1.0. If it shows locally built images, check that you are in the release bundle rather than the source repository.

Full profile with Phoenix#

To add Phoenix tracing to the same stack, run:

docker compose --profile full up -d

The UI is at http://localhost, the backend health endpoint at http://localhost:8000/health, and Phoenix at http://localhost:6006 when enabled. These local HTTP settings are for private evaluation. For shared access, configure HTTPS and authentication as described in Configure and validate.

Additional Setup#

The default CODE_GENERATOR_TOOL=auto path requires a valid Cursor credential for contract-first generation. Save the Cursor key in Settings when permitted or configure CURSOR_API_KEY and MODEL_CURSOR=auto in the deployment, as required by the v1.1.0 runtime. Use CODE_GENERATOR_TOOL=llm only when you intentionally select the separate inference-backed full-project mode. A GitHub token is optional for authenticated MCP lookups; fine-tuning needs separately configured NeMo Microservices endpoints.

Installation Verification#

Run the following checks from the release bundle directory.

  1. Check the stack and health endpoint:

    docker compose --profile mcp ps
curl -fsS http://localhost:8000/health

  2. Open http://localhost and confirm the frontend loads.

  3. In Settings, verify that the effective inference profile can reach its chat and embedding models. For the default auto generation path, validate the Cursor key before starting a workflow.

  4. Follow Generate your first POC, then restart the stack and confirm the POC remains visible and downloadable.

A healthy /health response proves service health; the POC and restart checks exercise the workflow and persistence.

Troubleshoot the Installation#

If launch preflight fails, inspect the missing variable named in its output and compare .env with the bundle’s env.example. If an image pull fails, check the NGC organization, login, image tag, and host architecture. See Troubleshooting for symptom-specific checks.

Next Steps#

After the first successful health check, validate the user workflow and backup plan.