LangChain: evidence-graph GraphRAG lifts multi-source agent accuracy from 10% to 52%
JeremyCMorgan · x · 2026-11-11
LangChain's new blog post explains building evidence-grounded agent memory with its open-source OpenWiki and evidence graphs.
Key results
- Using a compiled wiki as a memory layer made an agent 5x more accurate on cross-source questions than grep-only retrieval; for simple lookups, the memory-free agent was just as accurate and faster.
- Building the graph from document citations instead of LLM-extracted entities raised connected-question accuracy from 10% to 52%, with simple lookups barely changing.
- Adding the evidence graph doubled citation consistency vs. wiki alone, improved accuracy by 27%, and cut cost per correct answer by 58%.
Takeaway: GraphRAG value depends on question structure — persistent compiled memory plus an evidence graph pays off for repeated cross-source reasoning, not simple lookups.
More from coding & agent
- ZenML Pro launches always-on AI engineering team to keep agents in production — strickvl · 2026-11-11
- Latent Space tests NaiveAI's 2,100 tok/s coding model: fast, verbose, promising — Latent Space · 2026-11-11
- Anaconda's LLMOps guide: retrieval quality degrades independently of the model — anacondainc · 2026-11-11
- Dev demos agent with 'durable object brain' plus on-demand computer, local or remote — samgoodwin89 · 2026-11-11
- Right-sizing intelligence per role: frontier reasoning at the control plane, adversarial verifiers at the gate — clmt · 2026-11-11
- MCP Apps ship with your server and run in any AI app supporting the extension — ClaudeDevs · 2026-11-11