axoviq-ai/synthadoc (~534★) is an open-source knowledge-compilation engine: it turns raw documents into structured, local-first wikis and pitches itself as a transparent, human-readable alternative to RAG that is self-managed and self-improving without using any tools.

Why it's worth studying

• Actively engaged with the hard problems from the critiques: its docs cover adversarial review, claim provenance, page lifecycle management, and query caching — i.e. exactly the truth-maintenance and token-cost objections.
• "No tools" self-management is a useful contrast to the search-tool-backed scaling answer — it bets that disciplined compilation plus caching is enough.
• One of the more mature efforts (multiple releases, quick-start + design docs, web + CLI + Obsidian query paths).

Relevance to us

synthadoc is the best worked example of engineering answers to the critiques: adversarial review and claim-provenance features are concrete versions of the review-gated, citation-first fix, and query caching addresses the token-cost objection. Strong prior art to mine when we decide how much of the trust model our lint and ingest skills enforce.

Citations

1. synthadoc — https://github.com/axoviq-ai/synthadoc