--- title: How to build evidence-grounded agent memory with... description: LangChain tested its open-source OpenWiki compiler plus an evidence graph as a memory layer for agents answering cross-source questions. On 50 connected... canonical: https://daily.dev/posts/how-to-build-evidence-grounded-agent-memory-with-openwiki-and-graphs-3zh6ayksv twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: How to build evidence-grounded agent memory with OpenWiki and Graphs | daily.dev og:description: LangChain tested its open-source OpenWiki compiler plus an evidence graph as a memory layer for agents answering cross-source questions. On 50 connected... og:url: https://daily.dev/posts/how-to-build-evidence-grounded-agent-memory-with-openwiki-and-graphs-3zh6ayksv og:image: https://api.daily.dev/og/posts/3zH6AYkSV.png og:image:alt: How to build evidence-grounded agent memory with OpenWiki and Graphs og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # How to build evidence-grounded agent memory with OpenWiki and Graphs **[LangChain](https://daily.dev/sources/langchain)** · 10 min read · 0 upvotes · 0 comments ## Summary LangChain tested its open-source OpenWiki compiler plus an evidence graph as a memory layer for agents answering cross-source questions. On 50 connected questions over a synthetic 111-document insurance library, a wiki-equipped agent hit 0.52 accuracy versus 0.10 for keyword search alone, roughly a 5x gain. Adding an evidence graph of typed edges (cite, refers_to, supersedes) further raised accuracy 27%, doubled citation consistency, cut median answer time 40%, and lowered cost per correct answer 58%. For 80 simple lookup questions, search alone matched the wiki on accuracy and cost while being faster. Compiling the 111-document library cost $4.15, with new documents added for about $0.09 each. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Questions this post answers ### How much more accurate is an agent with a compiled wiki versus keyword search alone on questions that span multiple documents? On connected questions requiring reasoning across sources, an agent using OpenWiki's compiled wiki scored 0.52 accuracy versus 0.10 for keyword search alone, roughly a 5x improvement. This was measured on 50 connected questions over a synthetic 111-document insurance library with a 972-fact ledger used for grading. Cost per correct answer also fell from about $0.55 to $0.19. _Weighing a memory layer for your agentic RAG pipeline? daily.dev surfaces benchmarks like this as they emerge._ ### Does adding an evidence graph on top of a wiki actually improve agent citation consistency? Yes, adding a lightweight evidence graph of typed edges (cite, refers_to, supersedes) on top of OpenWiki's compiled wiki doubled citation consistency compared to the wiki alone, while also raising accuracy by 27%, cutting median answer time by 40%, and reducing cost per correct answer by 58%. The graph took under a second to build from existing citations with no extra model calls. _Building reproducible, auditable agent retrieval? daily.dev tracks techniques like evidence graphs for grounding citations._ ### Is it worth building a compiled wiki memory layer for simple single-document lookup questions? No, for simple single-passage lookups, search alone was just as accurate as a compiled wiki (0.84 vs 0.83 accuracy) and cheaper and faster, with latency of 9 seconds versus 12 seconds for the wiki. Compiled wiki memory pays off mainly for connected questions requiring cross-document reasoning, where repeated reuse offsets ingestion and maintenance costs. _Deciding where agent memory is worth the overhead? daily.dev helps engineers compare retrieval architecture trade-offs._ ## Similar posts on daily.dev - [Evaluating OpenWiki with WikiBench](https://daily.dev/posts/evaluating-openwiki-with-wikibench-f3uvqnfba) · LangChain · 0 upvotes · 0 comments - [How to use AI agents better than 99% of people](https://daily.dev/posts/how-to-use-ai-agents-better-than-99-of-people-5tzhgrsk9) · System Design Newsletter · 1 upvotes · 0 comments - [Knowledge Graph as context for LLMs: demonstrating decisive RCA and faster production performance](https://daily.dev/posts/knowledge-graph-as-context-for-llms-demonstrating-decisive-rca-and-faster-production-performance-mhm1hbjmi) · Grafana Labs · 7 upvotes · 1 comments --- Tags: [#ai-agents](https://daily.dev/tags/ai-agents), [#rag](https://daily.dev/tags/rag), [#langchain](https://daily.dev/tags/langchain) [View this post on daily.dev](https://daily.dev/posts/how-to-build-evidence-grounded-agent-memory-with-openwiki-and-graphs-3zh6ayksv) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"How to build evidence-grounded agent memory with OpenWiki and Graphs","url":"https://daily.dev/posts/how-to-build-evidence-grounded-agent-memory-with-openwiki-and-graphs-3zh6ayksv","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/how-to-build-evidence-grounded-agent-memory-with-openwiki-and-graphs-3zh6ayksv"},"datePublished":"2026-10-28T17:02:44.438Z","dateModified":"2026-10-28T17:22:24.473Z","description":"LangChain tested its open-source OpenWiki compiler plus an evidence graph as a memory layer for agents answering cross-source questions. On 50 connected...","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/55651d42f43b9a421f54192a8a5bd5f5?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/55651d42f43b9a421f54192a8a5bd5f5?_a=AQAEuop","isAccessibleForFree":true,"articleSection":"LangChain","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"LangChain","logo":"https://media.daily.dev/image/upload/t_logo,f_auto/v1/logos/0f4f2629e0724a849823f8cd0d913e13","url":"https://daily.dev/sources/langchain"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/how-to-build-evidence-grounded-agent-memory-with-openwiki-and-graphs-3zh6ayksv","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":0},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"ai-agents,rag,langchain","timeRequired":"PT10M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"LangChain","item":"https://daily.dev/sources/langchain"},{"@type":"ListItem","position":3,"name":"How to build evidence-grounded agent memory with OpenWiki and Graphs"}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/how-to-build-evidence-grounded-agent-memory-with-openwiki-and-graphs-3zh6ayksv#faq","mainEntity":[{"@type":"Question","name":"How much more accurate is an agent with a compiled wiki versus keyword search alone on questions that span multiple documents?","acceptedAnswer":{"@type":"Answer","text":"On connected questions requiring reasoning across sources, an agent using OpenWiki's compiled wiki scored 0.52 accuracy versus 0.10 for keyword search alone, roughly a 5x improvement. This was measured on 50 connected questions over a synthetic 111-document insurance library with a 972-fact ledger used for grading. Cost per correct answer also fell from about $0.55 to $0.19. Weighing a memory layer for your agentic RAG pipeline? daily.dev surfaces benchmarks like this as they emerge."}},{"@type":"Question","name":"Does adding an evidence graph on top of a wiki actually improve agent citation consistency?","acceptedAnswer":{"@type":"Answer","text":"Yes, adding a lightweight evidence graph of typed edges (cite, refers_to, supersedes) on top of OpenWiki's compiled wiki doubled citation consistency compared to the wiki alone, while also raising accuracy by 27%, cutting median answer time by 40%, and reducing cost per correct answer by 58%. The graph took under a second to build from existing citations with no extra model calls. Building reproducible, auditable agent retrieval? daily.dev tracks techniques like evidence graphs for grounding citations."}},{"@type":"Question","name":"Is it worth building a compiled wiki memory layer for simple single-document lookup questions?","acceptedAnswer":{"@type":"Answer","text":"No, for simple single-passage lookups, search alone was just as accurate as a compiled wiki (0.84 vs 0.83 accuracy) and cheaper and faster, with latency of 9 seconds versus 12 seconds for the wiki. Compiled wiki memory pays off mainly for connected questions requiring cross-document reasoning, where repeated reuse offsets ingestion and maintenance costs. Deciding where agent memory is worth the overhead? daily.dev helps engineers compare retrieval architecture trade-offs."}}]} ```