Skip to content

Releases: feather-store/feather

v0.5.0 — Context Graph + Living Context Engine

28 Feb 10:22

Choose a tag to compare

What's New in v0.5.0

Context Graph

  • Typed weighted edgesEdge struct {target_id, rel_type, weight} replaces flat links list
  • Reverse indexget_incoming(id) for instant reverse-edge lookups, rebuilt on load
  • db.link(from_id, to_id, rel_type, weight) — create typed directed edges between records
  • db.auto_link(modality, threshold, rel_type) — automatically link similar vectors by cosine threshold
  • 9 built-in relationship types via RelType: related_to, caused_by, supports, part_of, derived_from, contradicts, precedes, references, multimodal_of

Living Context Engine

  • db.context_chain(query, k, hops, modality) — vector search + n-hop BFS graph expansion in one call, returns ContextChainResult with scored nodes and traversed edges
  • Adaptive decaystickiness = 1 + log(1 + recall_count) makes frequently accessed records resist temporal decay

Graph Visualizer

  • visualize(db, output_path) — exports a fully self-contained D3 force-graph HTML (no CDN, works offline)
  • export_graph_json(namespace_filter, entity_filter) — D3/Cytoscape-compatible JSON export
  • Dangling edge safety — edges referencing excluded nodes are automatically filtered

Inspector

  • feather_inspector.py — local HTTP inspector with force graph view, PCA embedding scatter, similarity search, edit and delete

New Methods

  • db.get_vector(id, modality) — retrieve raw vector as NumPy array
  • db.get_all_ids(modality) — list all IDs in a modality index
  • db.update_metadata(id, meta) — update metadata without touching the HNSW index
  • db.update_importance(id, importance) — adjust importance score in-place

File Format

  • Bumped to v5 — adds typed edges with rel_type + weight; v3/v4 files load transparently

Install

pip install feather-db
cargo install feather-db-cli

Links