Quick start¶

Runs the full stack with Docker Compose. Everything below also works in PowerShell on Windows with Docker Desktop.

1. Configure¶

Copy config/profiles.example.yaml to config/profiles.yaml and set each profile's llm.summarizer endpoint — or put LOCAL_LLM_BASE_URL / LOCAL_LLM_API_KEY in .env (copy .env.example).

llm:
  summarizer:
    base_url: ${LOCAL_LLM_BASE_URL:-http://localhost:8080/v1}
    api_key: ${LOCAL_LLM_API_KEY:-}
    model: ${LOCAL_LLM_MODEL:-local}

Any OpenAI-compatible chat-completions endpoint works (llama.cpp llama-server, vLLM, Ollama, OpenAI). See LLM endpoints.

Config file encoding

profiles.yaml must be saved as UTF-8 and copied to the server with cp/scp — not PowerShell Get-Content/Set-Content, which can corrupt non-ASCII text and cause UnicodeDecodeError.

2. Start¶

docker compose up --build -d

The log shows the LLM health check (GET {base_url}/models). A 401 there is reported but tolerated.

3. Watch the first stories arrive¶

Open http://localhost:8000. Until the first story is fully generated the feed says "Preparing your first stories..." with live counts and any blocker (endpoint not set, unreachable, HTTP 401, paused). Stories appear one at a time as they publish ("N new stories — click to load").

Settings > Pipeline shows every story still in progress.

4. Start over¶

docker compose down -v   # drops the database volume too

Useful commands¶

curl -X POST localhost:8000/api/admin/refresh   # fetch all feeds now
docker compose logs -f app                     # watch ingestion and generation
docker compose down                            # stop (keep the database)

Next: first-run checklist · deploy on a NAS · configuration