Skip to content

Demo without an LLM

Two ways to see the app working without configuring a model: a local throwaway demo (make demo / demo.cmd) and a full Docker end-to-end run (scripts/docker_e2e.py).

Local demo (make demo)

Needs Python 3.12+. Creates a throwaway database under .demo/, seeds the example profiles, loads 66 fixture articles over a mock transport (no network), extracts, clusters, and generates briefs/reports with a deterministic in-process fake LLM. Card images are generated locally too.

make demo                      # http://localhost:8000
make demo PORT=8001            # different port
make demo DEMO_ARGS="--fake-llm-delay 3"   # slow answers: watch generating state

demo.cmd                  :: http://localhost:8000, opens your browser
demo.cmd --port 8001 --no-open

demo.cmd finds Python 3.12+ via py/python, creates .venv, installs dependencies on first run (reinstalls only when pyproject.toml changes), then starts the demo. python scripts/demo.py is the same launcher and works anywhere.

Demo extras:

  • A fake alert receiver (webhook + ntfy + SMTP) at http://localhost:8000/demo/sink; market-monitor alert rules are pointed at it and a few "live" items are ingested.
  • ?cards=lead_only on a feed URL previews lead-only cards.
  • ?no_image=logo|text_only|gradient previews the no-image fallbacks.
  • --report-mode off, --fake-chat-mode <mode> (stream_error, invented_number, stall, ...) exercise failure states.

Use a real model instead: set LOCAL_LLM_BASE_URL / LOCAL_LLM_API_KEY and run make demo DEMO_ARGS="--real-llm --model <model>" (Windows: .\demo.cmd --real-llm --model <model>). It generates briefs+reports for the newest 10 stories at startup.

Docker e2e (scripts/docker_e2e.py)

A fresh stack (own compose project discover-e2e, docker-compose.yml + docker-compose.fake-llm.yml) with fixture feeds, a fake LLM HTTP server and fake web-search providers. It waits for published stories, asserts the publish gate holds, prints a timeline and tears down. Stdlib only; works on Windows.

python scripts/docker_e2e.py --keep      # keep the stack running, then open :8000
# later: docker compose -p discover-e2e down -v

.venv\Scripts\python scripts\docker_e2e.py --keep

Options: --timeout 900, --out timeline.txt.