The Last Directory: a local ghost story with a paper trail

Description and use

The project shows how a local open‑weight model can generate short ghost stories from a fictional archive while providing a verifiable provenance trail.

Problem

How to generate short ghost‑story scenes from a fictional file index while keeping a verifiable paper trail of source evidence, without accessing real user data or external services.

Outcome

The project demonstrates that a local, open‑weight model can generate narrative content from a fictional archive while providing a verifiable paper trail of source evidence, enabling inspection and reproducibility without external inference services.

Workflow

• 1. Install runtime and start model with python3 setup_runtime.py sh start_model.sh
• 2. Run the application with python3 haunt.py
• 3. Open the web UI at http://127.0.0.1:8763
• 4. Select a scene (e.g., Station Nine) and click "Explore recorded inference"
• 5. View highlighted source entries and click "Save provenance" to download the provenance bundle

Evidence

The Last Directory turns a fictional file index into three short ghost‑story scenes, using a small open‑weight model running locally.

Source

The public site replays three saved outputs from actual local Spark runs. Its banner says RECORDED DEMO · NO ONLINE INFERENCE.

Source

The core is Spark‑X2.6‑4B, using the publisher's Q4_K_M GGUF with llama.cpp b11669 on an Apple M5 with 16GiB memory.

Source