Skip to main content

2. Quickstart: zero to a running dashboard (7 minutes)

Tutorial: docs/getting-started/quickstart.mdx · Highest-value piece: anyone who cannot get it running never sees the rest.

Goal: from a clean machine to the signed-in hub with the Business and Implementation portals.

Before recording: clean machine or clean Docker state, ollama pull already done once so the video does not wait on a download (say so), throwaway OPENAI_API_KEY in .env, ports free.

Chapters​

  • 0:00 What you will build
  • 0:30 Prerequisites
  • 1:30 Pull the model
  • 2:00 Clone and configure
  • 3:00 Start the backend
  • 4:30 Start the dashboard
  • 5:15 Sign in
  • 6:15 Pull the job image
  • 6:40 Next steps

0:00 What you will build [CHAPTER]​

SCREEN: Version card ("Tested on commit <sha>"), then the end state: the hub with the Business and Implementation portals.

VO: In this video you will run the complete OpenJCockpit stack on your own machine and sign in to the AI Orchestration Dashboard. At the end, you will see the hub with the Business and Implementation portals.

This demo uses a local model, qwen3.6 27b, through Ollama. Output from a local model varies from run to run, so yours may differ in detail.

0:30 Prerequisites [CHAPTER]​

SCREEN: Checklist appears, one item per line, as the VO mentions it.

VO: You need Docker with Docker Compose. You need Ollama installed on the host, not in Docker, and enough memory for a large model. You need Git, to clone two repositories side by side.

And you need an OpenAI API key. This surprises people: it is mandatory to boot the stack, because the best and cheapest model aliases and the embeddings are still OpenAI-hosted. Use a throwaway key for tutorials.

Finally, these host ports must be free: 4000, 4403 to 4406, 4100, 8080, 8091, 9080 and 8181.

1:30 Pull the model [CHAPTER]​

SCREEN: Terminal, large font.

ollama pull qwen3.6:27b
ollama list

VO: First, pull the model and check it is listed. Keep the Ollama daemon running for as long as the stack is up. I already pulled it before recording, so this is quick.

SCREEN: Checkpoint overlay: "ollama list shows qwen3.6:27b".

2:00 Clone and configure [CHAPTER]​

git clone https://github.com/OpenJCockpit/OpenJCockpit.git
git clone https://github.com/OpenJCockpit/OpenJCockpit-frontend.git
cd OpenJCockpit
cp .env.example .env

VO: Clone both repositories next to each other, then copy the example environment file.

SCREEN: Open .env in an editor. Only the OPENAI_API_KEY line is visible, with the value blurred or a dummy.

VO: Set your OpenAI key. Every other variable in the file is commented out and has a working local default. Never commit this file.

3:00 Start the backend [CHAPTER]​

docker compose up --build

SCREEN: Build output scrolls. [SPEED-UP] the build. Pause on the one-shot local-seed container exiting with code 0.

VO: Now start the backend. The first build takes a while, so I sped this up.

Watch for the local-seed container. It runs once, prepares the Keycloak roles, the local users and the LiteLLM keys, and then exits. It is safe to run again.

SCREEN: Checkpoint overlay: "Build finished, local-seed exited successfully". Then a caution box: "Health is not readiness for inference".

VO: One warning. The health endpoint reports up even when Ollama is unreachable. A missing model only shows up when the first LLM call fails. If that happens, the troubleshooting page has the answer.

4:30 Start the dashboard [CHAPTER]​

Second terminal:

cd OpenJCockpit-frontend
docker compose up --build

VO: In a second terminal, start the dashboard. It is a separate application. It proxies API calls to the ai-control-service on port 9080 and signs in against Keycloak on port 8080.

5:15 Sign in [CHAPTER]​

SCREEN: Browser at http://localhost:4000, redirect to the login page. Lower third: "Seeded users: local development only." Sign in as tony.

VO: Open localhost 4000. You are redirected to the login page. Sign in with a seeded local user. I will use tony, who also has the admin role. These seeded users exist for local development only. Never use them anywhere else.

SCREEN: Hub appears. Checkpoint overlay: greeting with first name, top bar, Business and Implementation portals.

VO: And there is the hub: a greeting, a top bar, and the Business and Implementation portals.

6:15 Pull the job image [CHAPTER]​

docker pull eclipse-temurin:25-jre

VO: One last step. Workflow runs on a project use a container image that is not pulled automatically, so pull it once now.

6:40 What just happened and next steps [CHAPTER]​

SCREEN: Simple diagram: Keycloak, PostgreSQL, LiteLLM, OPA, Java services in Docker; dashboard and Ollama outside, talking over host ports.

VO: Docker Compose started identity, the database, the LiteLLM gateway, the policy engine, and the Java services. Ollama serves the model for every agent.

To stop: docker compose stop keeps your state. Docker compose down keeps the volumes. But docker compose down with the dash v flag deletes the realm, the databases and the keys. Only use it for a full reset.

Next up: your first spec-driven change on the demo project.

SCREEN: Outro card.

Production notes​

  • Show real snags only if they occurred while recording: OpenAI key missing, Ollama not running, port clash. Link to the troubleshooting page in the description.
  • Description chapters = the headings above.
  • VERIFY: port list, the local-seed behaviour and the eclipse-temurin:25-jre image on a clean machine, since the tutorial is still marked unverified.