Installation#

Install POC Factory from the public NVIDIA NGC images using Docker Compose or Kubernetes. Download matching deployment files below, then follow the detailed guide for your environment.

Prerequisites#

Check these requirements before selecting an installation method.

  1. Confirm host and runtime requirements in the Support matrix.

  2. Obtain NGC image pull access for nvcr.io/nvidia and an inference provider API key, or plan for user-provided inference profiles.

  3. Obtain a Cursor credential for default auto generation, or deliberately choose the separate llm generation mode.

  4. Plan persistent state, a stable encryption key, and database credentials. Shared services also need HTTPS and authentication.

The NGC catalog lists the containers publicly; image downloads still depend on your NGC permissions and product access.

Installation Methods#

Choose Docker Compose for a trusted host or Kubernetes for a managed shared service. Use matching deployment files and select all three application images from the same release. The examples below use the release selected for this documentation.

Recommended first installation. Run the released stack without installing local Python or Node.js.

  1. Download the Compose deployment bundle.

  2. Extract the archive and open the poc_factory/ directory.

  3. Follow Install with Docker Compose to configure .env, authenticate to NGC, start the stack, and verify health.

The bundle contains the release’s README.md, docker-compose.yml, env.example, and scripts/validate-compose-env.sh. It contains deployment files; application images are pulled from NGC.

For operators of a shared service. Prepare cluster secrets, storage, database, ingress, and TLS before installing.

  1. Download the Helm chart.

  2. Extract it to obtain the poc-factory/ chart directory.

  3. Follow Install on Kubernetes, including explicit nvidia image organization overrides for all three containers.

The packaged chart matches the documented release. Its production values are an example to adapt to your environment.

Installation Verification#

Check the running services and the user-facing workflow before calling the deployment ready.

  1. Verify that backend, frontend, and GitHub MCP proxy use the selected release tags or corresponding approved digests.

  2. Confirm services are healthy and the frontend loads.

  3. Validate chat and embedding roles in Settings and the Cursor credential for auto generation.

  4. Generate, download, and inspect a small POC; restart the service and confirm that the record, archive, and saved credentials remain usable.

The detailed commands are in each installation guide. Configure and validate explains the full acceptance checks and skipped validation gates.

Troubleshooting#

For image-pull errors, verify NGC access, the nvidia namespace, the selected image tag, and the host architecture. A launch preflight error names the missing deployment value; put it in .env for Compose and its preflight container. See Troubleshooting for symptom-specific checks.

Next Steps#

Once the service is healthy, Generate your first POC. For an existing deployment, follow Upgrade guide instead of initializing a new environment.