[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"branding":3,"analytics":7,"article-tokenmizer-swaps-flat-chat-history-for-a-knowledge-graph":10,"sections":34},{"siteName":4,"siteTagline":5,"publisherName":4,"contactEmail":6},"The Revision","Tech news, decoded.","editor@therevision.news",{"gaMeasurementId":8,"adsenseClientId":9},"G-ZW2MV82GYR","ca-pub-8533917693782264",{"article":11},{"id":12,"slug":13,"title":14,"dek":15,"body_md":16,"tags_json":17,"published_at":18,"created_at":19,"updated_at":20,"status":21,"review_note":22,"review_notes":23,"image_url":22,"persona_id":22,"persona_name":22,"section":24,"tags":25,"sources":29,"feedback":33,"feedback_at":22,"cost_usd":33,"total_tokens":33},4081,"tokenmizer-swaps-flat-chat-history-for-a-knowledge-graph","TokenMizer Swaps Flat Chat History for a Knowledge Graph","An open-source proxy replaces raw LLM transcripts at context limits with a structured graph that preserves decisions, task state, and file history.","Long AI sessions have a memory problem, and most fixes make it worse.\n\nTokenMizer is an open-source transparent proxy that intercepts LLM sessions before they hit context-window limits and replaces the raw transcript with a compact, structured summary — not a plain-text summary, but a serialized knowledge graph. The graph tracks 14 node types and 7 edge types across an 8-state lifecycle, meaning it records not just what was decided but why, what evidence supported it, and when it was superseded. Version 0.3.1 adds a production serving layer: SSE streaming, nine provider adapters, a monitoring dashboard, and Model Context Protocol tools so agents can checkpoint and resume sessions directly. Graph exports land in D3 JSON, interactive HTML, or Obsidian Canvas.\n\nThe core insight is that truncation and summarization both treat conversation history as flat text, which shreds exactly the information that makes a session resumable — rationales, task status, and which files changed. A graph preserves those relationships explicitly. In early testing on three synthetic sessions, graph extraction matched a plain-summary baseline on task recall (75.6%) and beat it on decision recall (85.0% vs. 70.0%) and file recall (100% vs. 91.7%), with resume blocks generated in under 530 ms.\n\nThe authors are admirably candid: n=3 is directional, not conclusive, and they flag ceiling effects and baseline weaknesses in the paper itself. Still, the structural argument holds regardless of the small sample — and releasing the benchmark runner and exact results file under MIT means the numbers are at least reproducible, which is more than most AI memory papers can say.","[\"ai\",\"llm\",\"open-source\",\"developer-tools\"]","2026-07-07T04:00:00.000Z","2026-07-07T16:46:31.096Z","2026-07-07T16:46:34.057Z","published",null,[],"ai",[24,26,27,28],"llm","open-source","developer-tools",[30],{"name":31,"url":32},"arXiv cs.AI","https:\u002F\u002Farxiv.org\u002Fabs\u002F2606.06337",0,{"sections":35},[36,40,45,50,55,60,65,70,75,79,84,88,93,98],{"name":37,"slug":24,"count":38,"latest_published_at":39},"AI",2590,"2026-07-16T04:00:00.000Z",{"name":41,"slug":42,"count":43,"latest_published_at":44},"Security","security",294,"2026-07-15T19:59:48.000Z",{"name":46,"slug":47,"count":48,"latest_published_at":49},"Deals","deals",179,"2026-06-29T20:02:07.000Z",{"name":51,"slug":52,"count":53,"latest_published_at":54},"Policy","policy",158,"2026-07-16T00:02:48.000Z",{"name":56,"slug":57,"count":58,"latest_published_at":59},"Hardware","hardware",122,"2026-07-14T19:46:26.000Z",{"name":61,"slug":62,"count":63,"latest_published_at":64},"Consumer Tech","consumer-tech",93,"2026-07-13T13:20:48.000Z",{"name":66,"slug":67,"count":68,"latest_published_at":69},"Software","software",70,"2026-07-13T19:52:25.000Z",{"name":71,"slug":72,"count":73,"latest_published_at":74},"Science","science",66,"2026-07-10T10:29:37.000Z",{"name":76,"slug":77,"count":78,"latest_published_at":18},"Dev Tools","dev-tools",59,{"name":80,"slug":81,"count":82,"latest_published_at":83},"Gaming","gaming",41,"2026-07-09T04:00:00.000Z",{"name":85,"slug":86,"count":82,"latest_published_at":87},"Startups","startups","2026-06-29T20:55:50.000Z",{"name":89,"slug":90,"count":91,"latest_published_at":92},"General","general",29,"2026-07-10T22:28:58.000Z",{"name":94,"slug":95,"count":96,"latest_published_at":97},"Reviews","reviews",20,"2026-06-24T12:00:01.000Z",{"name":99,"slug":100,"count":101,"latest_published_at":102},"How-To","how-to",6,"2026-06-16T09:00:00.000Z"]