Skip to content

Quickstart

What runs today: clone, install, capture an X tweet (and a few other sources), then recall them by meaning.

1. Install

Terminal window
git clone https://github.com/KhiipAI/khiip.git ~/projects/khiip
cd ~/projects/khiip
pip install -e ".[dev]"

2. Start the daemon

Terminal window
# First run downloads the bundled MiniLM-L6 ONNX model (~80MB, one time)
khiipd serve &

The daemon listens on 127.0.0.1:8478 by default and reads/writes ~/.config/khiip/, ~/.local/share/khiip/, and ~/khiip-vault/.

3. Capture across multiple sources

Different extractors, same vault:

Terminal window
khiipd capture https://x.com/karpathy/status/2040470801506541998 # X: Karpathy's "LLM knowledge base" idea-file
khiipd capture https://simonwillison.net/2025/Dec/31/the-year-in-llms/ # web: Simon Willison — the year in LLMs
khiipd capture https://en.wikipedia.org/wiki/Quipu # wiki: the Inca knot-record system Khiip is named for
khiipd capture https://www.youtube.com/watch?v=-TAowrw97-4 # YouTube: Tiago Forte — the AI Second Brain

4. Recall by meaning

Terminal window
khiipd recall "Andrej Karpathy's LLM knowledge wiki"
khiipd recall "building a second brain with AI"
khiipd recall "Inca knot record-keeping system"

Recall ranks by cosine similarity over the bundled MiniLM-L6 embeddings — zero LLM cost, works offline after the first model fetch.

5. Inspect and maintain

Terminal window
# Check vault <-> SQLite consistency (read-only)
khiipd validate
# Refetch a capture along independent dimensions
khiipd refetch 01JX9... # re-extract (creates a new superseding capture)
khiipd refetch 01JX9... --media # re-walk the media-fetcher registry in place
khiipd refetch 01JX9... --wayback # re-submit to the Wayback Machine in place

What you end up with

  • A SQLite index at ~/.local/share/khiip/index.db
  • Markdown captures with YAML frontmatter under ~/khiip-vault/captures/<source>/
  • Raw Source-tier bytes preserved under your configured data_root
  • Semantic recall over everything you’ve captured

Next steps