Praxis v0.3.0 adds the first Core intake layer for converting source material before it enters capture, chunking, search, SkillGraph memory, or agent workflows.
This release adds source-type detection, document/media conversion, parse-quality scoring, converter metadata, and warning propagation so Praxis can better track whether a source was cleanly extracted before it becomes reusable agent knowledge.
Highlights include:
- PDF, Office, text, image, audio, and video intake paths.
- Parse-quality scores and extraction warnings.
- Source-linked evidence units for converted content.
- Optional OCR, speech-to-text, keyframe, scene, diarization, word-alignment, and visual-embedding adapters.
- Intake metadata carried into captures, chunks, and praxis search --explain.
- Lightweight default install, with heavier media dependencies kept optional.
This keeps Praxis moving toward a safer agent knowledge layer: sources are now inspected, converted, scored, and traced before downstream retrieval or skill export.
Full Changelog: v0.2.0...v0.3.0