// Use when working on, running, stopping, verifying, debugging, or configuring the HomeGrown AI App Demo project, including its Docker Compose stack, FastAPI app, PostgreSQL database, LiteLLM proxy, Prompt Security settings, model cache, provider keys, or custom inference endpoints.
| name | homegrown-ai-app-demo |
| description | Use when working on, running, stopping, verifying, debugging, or configuring the HomeGrown AI App Demo project, including its Docker Compose stack, FastAPI app, PostgreSQL database, LiteLLM proxy, Prompt Security settings, model cache, provider keys, or custom inference endpoints. |
Use this skill for project-specific app operations in the HomeGrown AI App Demo repository. It captures the known Docker Compose workflow, verification checks, and inference configuration points for this repo.
docker-compose.yml, app/, and litellm/app, db, litellmhttp://localhost:9000http://localhost:9000/adminhttp://localhost:4000.env file and there never will be. All configuration goes in the environment: sections of docker-compose.yml, or through the Admin → Settings panel at runtime.docker-compose.yml.litellm/config.yaml.docker-compose.yml or saved at runtime via Admin → Settings → LLM API Keys — runtime-saved keys take precedence and trigger automatic model discovery.Prefer the bundled script for routine operations:
bash .codex/skills/homegrown-ai-app-demo/scripts/manage.sh status
bash .codex/skills/homegrown-ai-app-demo/scripts/manage.sh start
bash .codex/skills/homegrown-ai-app-demo/scripts/manage.sh stop
bash .codex/skills/homegrown-ai-app-demo/scripts/manage.sh verify
bash .codex/skills/homegrown-ai-app-demo/scripts/manage.sh refresh-models
Run the helper from anywhere inside the repo. If the repo cannot be discovered from the current directory or script location, set HOMEGROWN_AI_APP_DIR=/path/to/homegrown-ai-app-demo.
git status --short.docker compose up -d --build.POST /admin/refresh-models.docker compose ps, GET /health, GET /, GET /admin, and authenticated GET /models.Run docker compose stop, then verify docker compose ps shows no project containers running.
There are two ways to add models:
1. LiteLLM config (static) — add a model entry to litellm/config.yaml and set any required API key in the environment: section of docker-compose.yml. Restart litellm and app, then refresh the model cache.
2. Direct provider routing (dynamic) — save an API key for OpenAI, Anthropic, Google, Perplexity, or OpenRouter in Admin → Settings → LLM API Keys. The app queries the provider's /models endpoint and adds all available models to the picker automatically as provider/model-id (e.g. openai/gpt-4.1). These bypass LiteLLM entirely.
For custom OpenAI-compatible endpoints, add the base URL and model ID via litellm/config.yaml. Never hardcode API keys in repo files — use docker-compose.yml environment sections or the admin Settings panel.
For the project’s local OpenAI-compatible test server, use:
http://localhost:8081/v1http://host.docker.internal:8081/v1huggingface/Qwen3VL-8B-Instruct-F16LOCAL_OPENAI_BASE_URL, LOCAL_OPENAI_API_KEY, LOCAL_OPENAI_MODEL_IDSdocker compose logs --no-color --tail=120 appdocker compose logs --no-color --tail=160 litellmdocker compose logs --no-color --tail=120 dbcurl -sS http://localhost:9000/health | python3 -m json.toolcurl -sS http://localhost:4000/v1/models -H "Authorization: Bearer $(docker compose exec app printenv LITELLM_MASTER_KEY)"Before saying the app is running, stopped, fixed, or configured, run fresh verification in the current turn and report the actual output state. Use superpowers:verification-before-completion when making completion claims.