[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"article-memos-plugin-fixes-hermes-agent-memory-en":3,"tags-memos-plugin-fixes-hermes-agent-memory-en":27,"related-lang-memos-plugin-fixes-hermes-agent-memory-en":28,"related-posts-memos-plugin-fixes-hermes-agent-memory-en":32,"series-ai-agent-90d31b81-cb67-46e8-82b7-ea3410e6acac":69},{"id":4,"title":5,"content":6,"summary":7,"source":8,"source_url":9,"author":9,"image_url":10,"keywords":11,"language":17,"translated_content":9,"views":18,"is_premium":19,"created_at":20,"updated_at":20,"cover_image":10,"published_at":20,"rewrite_status":21,"rewrite_error":9,"rewritten_from_id":9,"slug":22,"category":23,"related_article_id":24,"status":25,"google_indexed_at":26,"x_posted_at":9,"tweet_text":9,"title_rewritten_at":9,"title_original":9,"key_takeaways":9,"topic_cluster_id":9,"embedding":9,"is_canonical_seed":19},"90d31b81-cb67-46e8-82b7-ea3410e6acac","Hermes Agent Hits a Memory Wall. MemOS Shows What Comes Next.","\u003Cp>\u003Ca href=\"https:\u002F\u002Fgithub.com\u002FNousResearch\u002Fhermes-agent\" target=\"_blank\" rel=\"noopener\">Hermes Agent\u003C\u002Fa>, the open-source self-improving AI runtime from Nous Research, has become a reference implementation for personal AI assistants. Yet beneath its 122k stars lies an architectural shortcoming that becomes apparent only at scale: memory is broken.\u003C\u002Fp>\n\n\u003Cp>The problem is deceptively simple. Hermes stores every conversation turn directly to SQLite and retrieves via text matching. This works fine for proof-of-concepts. But as usage accumulates—thousands of turns, months of interaction—three pathologies emerge that no keyword search can fix.\u003C\u002Fp>\n\n\u003Ch2>Why Append-Only Memory Fails\u003C\u002Fh2>\n\n\u003Cp>First, duplication explodes. The same fact gets recorded multiple times. Second, contradictions coexist without resolution. Consider: the system records \"user is dieting\" on March 1st and \"user gave up dieting\" on April 15th. Both persist as separate entries. A month later, when the agent retrieves memories about the user's habits, it gets both records and cannot determine which is current. Third, signal-to-noise collapses. After 5,000 turns, a keyword search for \"food\" returns fifty disconnected fragments, most irrelevant to the current conversation.\u003C\u002Fp>\n\u003Cfigure class=\"my-6\">\u003Cimg src=\"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1777385817127-1p2e.png\" alt=\"Hermes Agent Hits a Memory Wall. MemOS Shows What Comes Next.\" class=\"rounded-xl w-full\" loading=\"lazy\" \u002F>\u003C\u002Ffigure>\n\n\n\u003Cp>This isn't unique to Hermes. \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fopenclaw\u002Fopenclaw\" target=\"_blank\" rel=\"noopener\">OpenClaw\u003C\u002Fa> (365k stars, the previous market leader) has identical limitations. Even GPT's memory feature for custom GPTs faces the same retrieval decay. The agent industry's dirty secret is that memory management has remained stubbornly primitive.\u003C\u002Fp>\n\n\u003Ch2>MemOS: Memory as Infrastructure\u003C\u002Fh2>\n\n\u003Cp>\u003Ca href=\"https:\u002F\u002Fgithub.com\u002FMemTensor\u002FMemOS\" target=\"_blank\" rel=\"noopener\">MemOS\u003C\u002Fa>, an Apache-licensed open-source memory operating system, offers a different architecture. Its \u003Ca href=\"https:\u002F\u002Fgithub.com\u002FMemTensor\u002FMemOS\u002Ftree\u002Fmain\u002Fapps\u002Fmemos-local-plugin\" target=\"_blank\" rel=\"noopener\">Hermes local plugin\u003C\u002Fa> intercepts the memory write path with four stages: semantic chunking → LLM summarization → vector embedding → intelligent deduplication.\u003C\u002Fp>\n\n\u003Cp>The deduplication logic is the pivot point. Rather than text similarity matching, an LLM evaluates semantic relationships between the new fact and existing memories, deciding: is this a duplicate, an update, or genuinely new? The dieting scenario becomes a single evolving record with edit history. On retrieval, a hybrid search engine runs full-text and vector semantic search in parallel, then fuses results with diversity dedup, time decay, and relevance filtering. Pre-fetching uses the user's latest message to pull relevant memories before the conversation begins.\u003C\u002Fp>\n\n\u003Ch2>Stratified Model Allocation\u003C\u002Fh2>\n\n\u003Cp>Hermes uses one model for all memory tasks, leading to uneven quality. MemOS introduces a three-tier config: lightweight embedding model, mid-tier summarization, heavyweight skill-generation model. Crucially, it adds a gating layer—only repeatable, high-value patterns become skills. A fallback chain ensures degradation: skill model → summary model → Hermes native. This treats memory as a system to be governed, not a passive log to be appended.\u003C\u002Fp>\n\u003Cfigure class=\"my-6\">\u003Cimg src=\"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1777385802429-uq4g.png\" alt=\"Hermes Agent Hits a Memory Wall. MemOS Shows What Comes Next.\" class=\"rounded-xl w-full\" loading=\"lazy\" \u002F>\u003C\u002Ffigure>\n\n\n\u003Cp>That's the real insight. Agents at scale don't need faster memory retrieval; they need memory curation. Without it, the agent eventually drowns in its own history.\u003C\u002Fp>\n\n\u003Ch2>Local-First Privacy and Multi-Agent Patterns\u003C\u002Fh2>\n\n\u003Cp>MemOS runs entirely on-device—SQLite plus local embedding models, zero cloud dependency. For teams, it supports dual-layer collaboration: multiple isolated Hermes instances on one machine sharing a common skill pool, or hub-client architecture across machines where private data never leaves home. Only explicitly shared content is team-visible.\u003C\u002Fp>\n\n\u003Cp>Installation is a single curl line (prerequisites: Node ≥18, Python 3, Hermes already installed). The plugin auto-detects environment, downloads binaries, symlinks into Hermes, and starts a local memory dashboard at \u003Ccode>127.0.0.1:18901\u003C\u002Fcode>. The dashboard offers memory browse, search, skill management, tool logs, and analytics—all localhost-only, password-protected.\u003C\u002Fp>\n\n\u003Ch2>The Memory Ceiling\u003C\u002Fh2>\n\n\u003Cp>The tradeoff is real: MemOS models consume tokens on first use, with benefits only apparent over months. But the alternative—native Hermes memory decaying into noise—isn't acceptable for production agents. Compare this to \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fmem0ai\u002Fmem0\" target=\"_blank\" rel=\"noopener\">mem0\u003C\u002Fa>, which offers similar memory governance but cloud-hosted. MemOS is the local-first answer.\u003C\u002Fp>\n\n\u003Cp>This pattern—treating memory as a first-class infrastructure problem—will spread across every major agent runtime within twelve months. It's not optional anymore.\u003C\u002Fp>","Hermes Agent has amassed 122k GitHub stars, but its native memory system uses append-only SQLite with keyword retrieval—adequate for hundreds of turns, broken at scale. MemOS's local plugin solves this through semantic chunking, LLM deduplication, and hybrid search, treating memory as a governed system rather than a transcript. This reveals a critical inflection point: agent runtimes are shifting from prototype-grade to production-grade memory architecture.","oracore-original",null,"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1777385817127-1p2e.png",[12,13,14,15,16],"Hermes Agent","MemOS","AI Memory","MemTensor","Agent Framework","en",0,false,"2026-04-28T14:16:09.137916+00:00","done","memos-plugin-fixes-hermes-agent-memory-en","ai-agent","d7849e5f-3341-4741-b047-e317e0bd0225","published","2026-04-29T09:00:09.786+00:00",[],{"id":24,"slug":29,"title":30,"language":31},"memos-plugin-fixes-hermes-agent-memory-zh","AI 代理人的記憶危機：為什麼本地模型需要語義記憶引擎","zh",[33,39,45,51,57,63],{"id":34,"slug":35,"title":36,"cover_image":37,"image_url":37,"created_at":38,"category":23},"fda44d24-7baf-4d91-a7f9-bbfecae20a27","switch-ai-outputs-markdown-to-html-en","How to Switch AI Outputs from Markdown to HTML","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1778743249827-wmsr.png","2026-05-14T07:20:22.631724+00:00",{"id":40,"slug":41,"title":42,"cover_image":43,"image_url":43,"created_at":44,"category":23},"064275f5-4282-47c3-8e4a-60fe8ac99246","anthropic-cat-wu-proactive-ai-assistants-en","Anthropic’s Cat Wu on proactive AI assistants","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1778735465548-a92i.png","2026-05-14T05:10:31.723441+00:00",{"id":46,"slug":47,"title":48,"cover_image":49,"image_url":49,"created_at":50,"category":23},"423ac8ad-2886-42a9-8dd8-78e5d43a1574","how-to-run-hermes-agent-on-discord-en","How to Run Hermes Agent on Discord","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1778724656141-i30t.png","2026-05-14T02:10:35.727086+00:00",{"id":52,"slug":53,"title":54,"cover_image":55,"image_url":55,"created_at":56,"category":23},"776a562c-99a6-4a6b-93a0-9af40300f3f2","why-ragflow-is-the-right-open-source-rag-engine-to-self-host-en","Why RAGFlow is the right open-source RAG engine to self-host","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1778674254587-0pxn.png","2026-05-13T12:10:25.721583+00:00",{"id":58,"slug":59,"title":60,"cover_image":61,"image_url":61,"created_at":62,"category":23},"322ec8bc-61d3-4c80-bb9e-a19941e137c6","how-to-add-temporal-rag-in-production-en","How to Add Temporal RAG in Production","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1778667085221-0mox.png","2026-05-13T10:10:31.619892+00:00",{"id":64,"slug":65,"title":66,"cover_image":67,"image_url":67,"created_at":68,"category":23},"1c09aef7-24bc-4d3a-b6cb-426b1012f432","github-agentic-workflows-ai-github-actions-en","GitHub Agentic Workflows puts AI agents in Actions","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1778551887736-7b7l.png","2026-05-12T02:11:07.184824+00:00",[70,75,80,85,90,95,100,105,110,115],{"id":71,"slug":72,"title":73,"created_at":74},"03db8de8-8dc2-4ac1-9cf7-898782efbb1f","anthropic-claude-ai-agent-task-automation-en","Anthropic's Claude AI Agent: A New Era of Task Automation","2026-03-25T16:25:06.513026+00:00",{"id":76,"slug":77,"title":78,"created_at":79},"045d1abc-190d-4594-8c95-91e2a26f0c5a","googles-2026-ai-agent-report-decoded-en","Google’s 2026 AI Agent Report, Decoded","2026-03-26T11:15:23.046616+00:00",{"id":81,"slug":82,"title":83,"created_at":84},"e64aba21-254b-4f93-aa21-837484bb52ec","kimi-k25-review-stronger-still-not-legend-en","Kimi K2.5 review: stronger, still not a legend","2026-03-27T07:15:55.385951+00:00",{"id":86,"slug":87,"title":88,"created_at":89},"30dfb781-a1b2-4add-aebe-b3df40247c37","claude-code-controls-mac-desktop-en","Claude Code now controls your Mac desktop","2026-03-28T03:01:59.384091+00:00",{"id":91,"slug":92,"title":93,"created_at":94},"254405b6-7833-4800-8e13-f5196deefbe6","cloudflare-100x-faster-ai-agent-sandbox-en","Cloudflare’s 100x Faster AI Agent Sandbox","2026-03-28T03:09:44.356437+00:00",{"id":96,"slug":97,"title":98,"created_at":99},"04f29b7f-9b91-4306-89a7-97d725e6e1ba","openai-backs-isara-agent-swarm-bet-en","OpenAI backs Isara’s agent-swarm bet","2026-03-28T03:15:27.849766+00:00",{"id":101,"slug":102,"title":103,"created_at":104},"3b0bf479-e4ae-4703-9666-721a7e0cdb91","openai-plan-automated-ai-researcher-en","OpenAI’s plan for an automated AI researcher","2026-03-28T03:17:42.312819+00:00",{"id":106,"slug":107,"title":108,"created_at":109},"fe91bce0-b85d-4efa-a207-24ae9939c29f","harness-engineering-ai-agent-reliability-2026","Harness Engineering: From Bridle to Operating System, The Missing Link in AI Agent Reliability","2026-03-31T06:36:55.648751+00:00",{"id":111,"slug":112,"title":113,"created_at":114},"67dc66da-ca46-4aa5-970b-e997a39fe109","openai-codex-plugin-claude-code-en","OpenAI puts Codex inside Claude Code","2026-04-01T09:21:55.381386+00:00",{"id":116,"slug":117,"title":118,"created_at":119},"7a09007d-820f-43b3-8607-8ad1bfcb94c8","mcp-explained-from-prompts-to-production-en","MCP Explained: From Prompts to Production","2026-04-01T09:24:40.089177+00:00"]