# LangChain's OpenWiki and evidence graph raise agent accuracy on cross-document questions

> Original title: Building Evidence-Grounded Agent Memory with OpenWiki and Graphs

- Source: LangChain Blog
- Published: 2026-10-28T04:00:00.000Z
- Bad Influence: https://badinfluence.news/items/djezh18wz7vimirz5pn8l5wl3
- Original: https://www.langchain.com/blog/building-agent-memory-openwiki-graphs

## Summary

LangChain tested OpenWiki, its open-source wiki compiler, as a memory layer for agents on a 111-document synthetic insurance library. On connected questions, the wiki raised accuracy from 0.10 with keyword search to 0.52 and cut cost per correct answer from about $0.55 to $0.19. Citation overlap across runs fell to 0.25 on those questions, which LangChain says a further evidence graph was meant to address.