01 / The short version
What happened
Researchers tested OpenWiki, an open-source wiki compiler, as a persistent memory layer for agents navigating scattered enterprise knowledge. On connected questions requiring cross-source reasoning, OpenWiki improved accuracy roughly 5x compared to search alone, rising from 0.10 to 0.52. Adding an evidence graph improved accuracy a further 27% and doubled citation consistency. The publisher reports cost per correct answer dropped from about $0.55 to $0.19 for connected questions. Compilation of a 111-document insurance library cost $4.15, with updates costing about $0.09 per document. Simple lookup results were similar between search and wiki, with wiki latency increasing from 9 to 12 seconds.
See the exact references02 / Key takeaways
What you need to know
- 01
OpenWiki improved connected-question accuracy from 0.10 to 0.52 versus search alone.
- 02
Adding an evidence graph improved accuracy by a further 27% and doubled citation consistency.
- 03
Publisher reports cost per correct answer fell from about $0.55 to $0.19 for connected questions.
Keep in perspective
What to watch for
Results are based on a synthetic insurance document library and publisher-reported metrics without independent validation.
Go to the source
Exact references
These are the original pages used for this brief. Publisher claims are not independent evaluations.
01Primary source · LangChainRead the original announcementhttps://www.langchain.com/blog/building-agent-memory-openwiki-graphsAI-generated from the linked source. It can miss context; verify consequential details in the original. How the radar works