GitHub Repository

Karpathy-style LLM Wiki Desktop App: hydrate raw docs into dense Markdown entries; retrieve by keyword + link traversal. No embeddings, no vector DB. Karpathy llm wiki, pdf-parser, knowledge-base, rag。https://arxiv.org/abs/2605.15184

12 starsTypeScript

Pith – A local-first desktop LLM wiki without vector DBs or embeddings

Name: Pith – A local-first desktop LLM wiki without vector DBs or embeddings
Availability: InStock
Author: a569171010

by a569171010·Jun 25, 2026·3 points·0 comments

Visit Project View on HN

AI Analysis

●●●BangerBig BrainBold BetSolve My Problem

Obsidian-compatible wiki skips vector DBs for LLM-hydrated Markdown and keyword traversal.

Strengths

•Hydration compresses docs to 30% size, reducing context window costs during retrieval.
•Plain Markdown storage avoids lock-in and works with Git or Obsidian natively.
•Keyword search and link traversal avoid embedding drift and vector DB overhead.

Weaknesses

•Keyword search on summaries may miss details not captured during hydration.
•Windows support is usable but not CI-covered, risking stability for many users.