v0.8.2

• Removed the ecosystem cross-link header so the README opens on the project
  heading.
• Added a chat-context comparison image and a process-flow diagram, and moved
  the graph visualizer demo down to "What It Does".
• Pointed README images at absolute raw URLs so they render on the PyPI project
  page as well as on GitHub.

v0.8.1

• Reworked the README: moved "What It Does" and "How Memory Enters The Graph"
  after Quick Start, documented the app-session (no API key) candidates flow,
  and simplified the install commands.
• Added llm.txt, an llms.txt-style summary of the project principles and
  entry points for AI assistants.

knowledge-worker 0.8.0

knowledge-worker 0.8.0

This release makes JSON-LD the default storage direction for knowledge-worker.

Highlights

• JSON-LD is now the canonical local graph storage format.
• JSONL remains the append-only history layer for review, eval, analyzer, and replay records.
• mykg export now defaults to canonical JSON-LD.
• mykg export --jsonld remains available as the explicit JSON-LD form.
• mykg export --ttl remains the Turtle/RDF interchange path.
• Legacy JSON graph files remain readable for migration.
• Kuzu is deferred as a future optional local read/query backend.
• graphify.net is deferred as a future publishing or interchange target.

Docs And Demo

• Added the storage decision record: docs/STORAGE_DECISION.md.
• Updated the roadmap, specification, design notes, principles, README, benchmarks, and contributor docs.
• Made examples/demo_graph.jsonld the canonical public demo graph.

Verification

• python3 -m unittest passed: 71 tests.
• Demo graph provenance check passed with provenance violations: 0.
• JSON-LD export validates as JSON.
• Turtle/RDF round-trip remains valid.
• Wheel builds as knowledge_worker-0.8.0.

knowledge-worker 0.7.0

Summary

knowledge-worker 0.7.0 adds the first Deep Dive workflow: a pre-ingest workspace for turning source material into reviewable artifacts, validation reports, and canonical graph candidates before anything becomes durable memory.

The release keeps the core product boundary intact:

The model proposes. Artifacts expose reasoning. Provenance verifies. Human review promotes.

What's New

• Added mykg deep-dive <source.md> --out-dir <workspace> to generate a pre-ingest workspace.
• Added mykg deep-dive inspect <workspace> to review artifacts, candidate counts, validation status, and next steps.
• Added mykg deep-dive add-to-graph <workspace> as a wrapper over the existing ingest validation/review/merge path.
• Added workspace outputs:
  • manifest.json
  • artifact-plan.json
  • generated Markdown artifacts
  • validation-report.json
  • artifact-graph.json
  • canonical *.candidates.json
• Added docs/DEEP_DIVE_INTERACTION_MODEL.md to define generate, inspect, challenge, add-to-graph, approve, and don't-ingest-yet semantics.
• Expanded README docs for how candidates become durable graph memory.
• Updated the roadmap with the v0.7.0 workspace release and the next memory analyzer direction.
• Added CHANGELOG.md.
• Added broader CLI regression coverage across existing commands and the new deep-dive command surface.
• Added GitHub Actions CI for PRs and main-branch pushes across Python 3.10, 3.11, 3.12, and 3.13.

Safety Model

deep-dive generation does not mutate MYGRAPH_PATH.

Graph mutation still goes through the existing ingest lifecycle:

candidates.json -> validate -> review -> merge accepted nodes/edges

deep-dive add-to-graph is a convenience wrapper over ingest, not a second graph merge path.

Example

mykg deep-dive notes.md --out-dir /tmp/deepdive-notes
mykg deep-dive inspect /tmp/deepdive-notes
mykg deep-dive add-to-graph /tmp/deepdive-notes

Validation

Local release checks passed:

Ran 68 tests
OK

CI now runs the unittest suite and demo graph provenance check on pull requests and main.

Follow-Up

The next roadmap layer is a memory analyzer that combines audit, context export, discovery, and deep-dive outputs into one review artifact: what the graph knows, what it can prove, what is weak or stale, what should be exported, and what should be reviewed before promotion.

0.6.2

Expand PyPI classifiers and keywords (AI-memory framing). Metadata only, no code changes.

0.6.1

Fix demo gif rendering on the PyPI project page (absolute asset URL). No code changes.

0.6.0

First PyPI release: pip packaging, OIDC publish workflow, Windows portability, /ingest-notes skill.