Skip to content

Coyote v0.5.1 — MVP

Latest
Latest

Choose a tag to compare

View all tags
@captaininfo captaininfo released this 06 Sep 04:27
· 9 commits to main since this release
v0.5.1
4b33990

A free, open-source, local-first tool that turns what you read and write online into a private, semantic knowledge graph on your own machine — a searchable record of your self-directed, informal learning, explorable and queryable with a local LLM. Nothing leaves your computer.

v0.5.1 makes first-time install actually work and adds contributor docs. The application is the same MVP as v0.5.0.

Get it

  • Windows: download Source code (zip) below.
  • macOS / Linux: download Source code (tar.gz) below.
  • Prefer git: git clone https://github.com/captaininfo/coyote.git && cd coyote && git checkout v0.5.1

You need both the app (a source archive above) and the Mozilla-signed Firefox extension coyote_browser_extension-2.0.0.xpi (attached below) — nothing is captured without the extension.

Requires: Docker (Desktop, or Engine + Compose on Linux), Firefox 140+, and Python 3.10+.

→ Full step-by-step install instructions are in the README. Start there. First run downloads and builds several GB and can take a while on a slow connection — the README explains the low-bandwidth path (Start Core first).

Fixed since v0.5.0

  • Fresh install now populates the graph on first run — the auto-generated Neo4j password reaches Core, so no manual Configure step is needed (v0.5.0 could come up "Online" yet silently store 0 nodes).
  • Apple-Silicon (arm64) image builds now work.
  • One-click "Start All" reliably pulls the local model and no longer deadlocks or gets killed on slow cold builds.
  • Clearer failures — compose errors surface in the status line and logs, and chat errors render on the turn they happen.

What's in this MVP

  • Browser capture (pages visited + links clicked) via a Mozilla-signed Firefox extension
  • Hypothes.is annotation import
  • A deterministic, auditable local NLP pipeline: KeyBERT topics, spaCy entities, pooled full-document embeddings, Wikidata concept linking
  • A Neo4j knowledge graph you can inspect directly
  • A GraphRAG chat assistant running a local model (qwen2.5-coder:3b by default)
  • Two runnable demos in demos/: source inference (which of your reading likely informed a note) and knowledge-shape (a coverage map of the concepts your reading covers)

Scope — not in this release

  • The in-browser search box (Purpose/SearchTerms capture) is disabled — see the README "Not in this release" note.
  • Capture is Firefox-only.
  • The Learning Insights dashboard panel is a "coming post-MVP" placeholder.

See the README and CLAUDE.md for the full, honest list of Known Issues.

This is an early MVP built in the open. Feedback, issues, and contributions are very welcome.

Assets 3
Loading