Skip to content

Changelog

All notable changes to Odin (odin-engine) are documented here. This project follows Semantic Versioning: patch releases carry backward-compatible fixes, minor releases add backward-compatible features, and major releases may break the public API.

0.3.0 2026-08-10

ekizito96

Tooling to make AI coding agents write correct Odin.

Added

  • The odin command-line tool. odin init --skill installs the packaged AI-editor skill into .github/skills/odin/ and .claude/skills/odin/, where GitHub Copilot and Claude Code load it automatically. Run odin init --skill --force to refresh after upgrades.
  • The AI-editor skill is packaged in the wheel, so it is versioned with the code and works offline.
  • An AI Editor Setup guide and a served llms.txt index of the documentation for AI ingestion.

0.2.1 2026-08-10

ekizito96

First release from the public Odin-1 repository. This is a packaging and documentation release; the engine code is identical to 0.2.0, so there are no public API changes.

Changed

  • Project metadata now points at the public GitHub repository and the documentation site at odin.developers.prescottdata.io.
  • Trimmed runtime requirements to the packages the engine actually imports: torch, python-arango, numpy, scipy, networkx, scikit-learn, and gremlinpython.

Added

  • Full documentation site built with MkDocs Material, covering concepts, guides, examples, and reference.
  • CHANGELOG.md, CONTRIBUTING.md, and CODE_OF_CONDUCT.md.
  • GitHub Actions for CI, PyPI publishing via trusted publishing, and documentation deployment.
  • The Odin brand-asset system (monogram, wordmark, and combo lockups).

0.2.0 2026-02-04

ekizito96

The baseline public release of the Odin engine on PyPI.

Added

  • Runtime schema introspection via SchemaInspector and inspect_arango_schema, giving agents live ArangoDB structure for writing valid queries.

Baseline features

  • The OdinEngine retrieval pipeline: Personalized PageRank, then beam search, then NPLL edge scoring, then aggregation into motifs, relation shares, and a triage score.
  • A self-managing NPLL lifecycle that trains from your graph on first run and persists the learned weights in ArangoDB.
  • Edge plausibility scoring via score_edge() and PPR-based anchor discovery via find_anchors().