Show HN: A Karpathy-style LLM wiki your agents maintain (Markdown and Git)
Karpathy's knowledge substrate idea, shipped. Markdown + Git + BM25 = agents that learn across sessions without vector DBs.
Why it matters
A working implementation of the LLM-native wiki pattern gaining attention in agent architecture circles—agents read from and write to a shared knowledge layer that persists across sessions, eliminating context re-pasting. Early indicator of how agent teams will manage institutional memory at scale.
The key facts
8 to knowMarkdown + Git as source of truth (not vector/graph DB)
BM25 + SQLite index; 85% recall@20 on 500-artifact benchmark
Draft-to-wiki promotion flow with human/agent review
Append-only JSONL fact logs per entity with synthesis worker
Open source (MIT); ships as WUPHF layer, agent-agnostic
Local-first (~/.wuphf/wiki/); users can git clone knowledge out
Deterministic fact IDs, canonical slugs, broken-link detection
Integrated /lookup slash command + MCP tool for cited retrieval
Go to the source
Hacker Newsgithub.com
Publisher excerpt: I shipped a wiki layer for AI agents that uses markdown + git as the source of truth, with a bleve (BM25) + SQLite index on top. No vector or graph db yet. It runs locally in ~/.wuphf/wiki/ and you can git clone it out if you want to take your knowledge with you. The shape is the one Karpathy has…