--- schema: 1 type: video title: >- CrabRAG: Why Automated Assistants Need Graph Memory, Not More Tokens — Stephen Chin, Neo4j description: >- Stephen Chin gave two copies of the same agent the same facts about his home network, one storing them as a vector database, the other as a graph. He asked both what was running end of life software exposed to the internet. The vector agent said it could not find specific details. The graph agent traced the connections and flagged his daughter's Minecraft server running an out of date OS, then found real open management ports he quietly patched after the demo. Same data, and only one of them could actually answer. That gap is the whole talk. Most assistants, OpenClaw included, keep their memory as markdown files, which is why Chin's agents burn over 100,000 tokens a round loading everything in case some of it matters. It holds at small scale and breaks at large scale, because similarity in vector space is not a real relationship, so multi hop questions hallucinate. A graph stores entities and the edges between them, seeds the search with vectors, then traverses, so answers come back precise, explainable, and auditable. And if you do not know graphs, Claude writes Cypher better than he does. Speaker info: - https://x.com/steveonjava - https://www.linkedin.com/in/steveonjava/ - https://www.oreilly.com/library/view/graphrag-the-definitive/9798341630147/ Timestamps: 0:00 - Meet Crab D and the agent memory problem 2:34 - Why markdown memory wastes tokens 4:43 - Skills are just markdown too 5:49 - Goose: memory as an MCP server 7:44 - Vector databases and why similarity is not a relationship 9:54 - Enter graphs: precise, explainable, auditable 11:38 - You do not need to be a graph expert, Claude writes Cypher 12:04 - The demo: a home lab digital twin, vector versus graph 13:23 - Live: finding end of life software on the network 15:30 - Live: finding exposed management ports 16:49 - Why large scale needs graph memory 18:05 - Resources: the GraphRAG book and GraphAcademy --- Watch along with the corrected transcript, summary, timestamps and resources on this talk's official AIE page: https://ai.engineer/talks/Q0VkgCyNVUg-crabrag-why-automated-assistants-need-graph-memory-not-more-tokens Join our upcoming AI Engineer conferences (see https://ai.engineer landing page): - AIEi Shanghai, November 5–6 — the first for meeting top Chinese labs and engineers - AIE CODE in SF, November 10–12 — the top AI Coding conference returns - AIEi Sydney, December 7–8 — the first AIEi run alongside NeurIPS 2026 in Sydney --- canonical: >- https://liquidgold.cc/channel/aidotengineer/explore/video/crabrag-why-automated-assistants-need-graph-memory-9b32e53f009f markdown: >- https://liquidgold.cc/channel/aidotengineer/video/crabrag-why-automated-assistants-need-graph-memory-9b32e53f009f.md llms: null parent: 'https://liquidgold.cc/channel/aidotengineer.md' channel: aidotengineer updated: '2026-10-24T06:54:32.177Z' language: null children: [] videoId: Q0VkgCyNVUg durationSeconds: 1242 publishedAt: '2026-07-22T18:30:02.000Z' searchChunks: - ordinal: 0 text: null startSeconds: 23 endSeconds: 160 watch: 'https://liquidgold.cc/watch/Q0VkgCyNVUg?t=23' heldReason: rights - ordinal: 1 text: null startSeconds: 171 endSeconds: 283 watch: 'https://liquidgold.cc/watch/Q0VkgCyNVUg?t=171' heldReason: rights - ordinal: 2 text: null startSeconds: 171 endSeconds: 283 watch: 'https://liquidgold.cc/watch/Q0VkgCyNVUg?t=171' heldReason: rights - ordinal: 3 text: null startSeconds: 293 endSeconds: 350 watch: 'https://liquidgold.cc/watch/Q0VkgCyNVUg?t=293' heldReason: rights - ordinal: 4 text: null startSeconds: 361 endSeconds: 464 watch: 'https://liquidgold.cc/watch/Q0VkgCyNVUg?t=361' heldReason: rights - ordinal: 5 text: null startSeconds: 361 endSeconds: 464 watch: 'https://liquidgold.cc/watch/Q0VkgCyNVUg?t=361' heldReason: rights - ordinal: 6 text: null startSeconds: 474 endSeconds: 598 watch: 'https://liquidgold.cc/watch/Q0VkgCyNVUg?t=474' heldReason: rights - ordinal: 7 text: null startSeconds: 474 endSeconds: 598 watch: 'https://liquidgold.cc/watch/Q0VkgCyNVUg?t=474' heldReason: rights - ordinal: 8 text: null startSeconds: 610 endSeconds: 700 watch: 'https://liquidgold.cc/watch/Q0VkgCyNVUg?t=610' heldReason: rights - ordinal: 9 text: null startSeconds: 610 endSeconds: 700 watch: 'https://liquidgold.cc/watch/Q0VkgCyNVUg?t=610' heldReason: rights - ordinal: 10 text: null startSeconds: 700 endSeconds: 711 watch: 'https://liquidgold.cc/watch/Q0VkgCyNVUg?t=700' heldReason: rights - ordinal: 11 text: null startSeconds: 733 endSeconds: 745 watch: 'https://liquidgold.cc/watch/Q0VkgCyNVUg?t=733' heldReason: rights - ordinal: 12 text: null startSeconds: 733 endSeconds: 745 watch: 'https://liquidgold.cc/watch/Q0VkgCyNVUg?t=733' heldReason: rights - ordinal: 13 text: null startSeconds: 816 endSeconds: 932 watch: 'https://liquidgold.cc/watch/Q0VkgCyNVUg?t=816' heldReason: rights - ordinal: 14 text: null startSeconds: 944 endSeconds: 1009 watch: 'https://liquidgold.cc/watch/Q0VkgCyNVUg?t=944' heldReason: rights - ordinal: 15 text: null startSeconds: 1020 endSeconds: 1085 watch: 'https://liquidgold.cc/watch/Q0VkgCyNVUg?t=1020' heldReason: rights - ordinal: 16 text: null startSeconds: 1085 endSeconds: 1242 watch: 'https://liquidgold.cc/watch/Q0VkgCyNVUg?t=1085' heldReason: rights --- # CrabRAG: Why Automated Assistants Need Graph Memory, Not More Tokens — Stephen Chin, Neo4j Stephen Chin gave two copies of the same agent the same facts about his home network, one storing them as a vector database, the other as a graph. He asked both what was running end of life software exposed to the internet. The vector agent said it could not find specific details. The graph agent traced the connections and flagged his daughter's Minecraft server running an out of date OS, then found real open management ports he quietly patched after the demo. Same data, and only one of them could actually answer. That gap is the whole talk. Most assistants, OpenClaw included, keep their memory as markdown files, which is why Chin's agents burn over 100,000 tokens a round loading everything in case some of it matters. It holds at small scale and breaks at large scale, because similarity in vector space is not a real relationship, so multi hop questions hallucinate. A graph stores entities and the edges between them, seeds the search with vectors, then traverses, so answers come back precise, explainable, and auditable. And if you do not know graphs, Claude writes Cypher better than he does. Speaker info: - https://x.com/steveonjava - https://www.linkedin.com/in/steveonjava/ - https://www.oreilly.com/library/view/graphrag-the-definitive/9798341630147/ Timestamps: 0:00 - Meet Crab D and the agent memory problem 2:34 - Why markdown memory wastes tokens 4:43 - Skills are just markdown too 5:49 - Goose: memory as an MCP server 7:44 - Vector databases and why similarity is not a relationship 9:54 - Enter graphs: precise, explainable, auditable 11:38 - You do not need to be a graph expert, Claude writes Cypher 12:04 - The demo: a home lab digital twin, vector versus graph 13:23 - Live: finding end of life software on the network 15:30 - Live: finding exposed management ports 16:49 - Why large scale needs graph memory 18:05 - Resources: the GraphRAG book and GraphAcademy --- Watch along with the corrected transcript, summary, timestamps and resources on this talk's official AIE page: https://ai.engineer/talks/Q0VkgCyNVUg-crabrag-why-automated-assistants-need-graph-memory-not-more-tokens Join our upcoming AI Engineer conferences (see https://ai.engineer landing page): - AIEi Shanghai, November 5–6 — the first for meeting top Chinese labs and engineers - AIE CODE in SF, November 10–12 — the top AI Coding conference returns - AIEi Sydney, December 7–8 — the first AIEi run alongside NeurIPS 2026 in Sydney --- Canonical: https://liquidgold.cc/channel/aidotengineer/explore/video/crabrag-why-automated-assistants-need-graph-memory-9b32e53f009f · Channel: [AI Engineer](https://liquidgold.cc/channel/aidotengineer.md) > Search chunk text is held until public-source provenance is verified. ## Document chunks > Missing: no stored document chunks are available.