{"type":"session","version":3,"id":"2aaba15e-28c3-467b-9cdf-f39c7bd803ef","timestamp":"2026-05-24T02:07:14.574Z","cwd":"/root/.openclaw/workspace"}
{"type":"model_change","id":"d5990c3f","parentId":null,"timestamp":"2026-05-24T02:07:14.576Z","provider":"openai_proxy","modelId":"gpt-5.4-mini"}
{"type":"thinking_level_change","id":"44a12b39","parentId":"d5990c3f","timestamp":"2026-05-24T02:07:14.576Z","thinkingLevel":"off"}
{"type":"custom","customType":"model-snapshot","data":{"timestamp":1779588434577,"provider":"openai_proxy","modelApi":"openai-responses","modelId":"gpt-5.4-mini"},"id":"893ed199","parentId":"44a12b39","timestamp":"2026-05-24T02:07:14.577Z"}
{"type":"message","id":"b155c468","parentId":"893ed199","timestamp":"2026-05-24T02:07:14.586Z","message":{"role":"user","content":[{"type":"text","text":"System: [2026-05-24 10:07:13 GMT+8] Slack DM from 李力: 有没有类似于hermes的新的这种项目出现\n\nConversation info (untrusted metadata):\n```json\n{\n  \"message_id\": \"1779588432.621559\",\n  \"reply_to_id\": \"1779588432.621559\",\n  \"sender_id\": \"U0A8ZCK7RJM\",\n  \"sender\": \"李力\",\n  \"timestamp\": \"Sun 2026-05-24 10:07 GMT+8\"\n}\n```\n\nSender (untrusted metadata):\n```json\n{\n  \"label\": \"李力 (U0A8ZCK7RJM)\",\n  \"id\": \"U0A8ZCK7RJM\",\n  \"name\": \"李力\"\n}\n```\n\n有没有类似于hermes的新的这种项目出现"}],"timestamp":1779588434583}}
{"type":"message","id":"d92d6fec","parentId":"b155c468","timestamp":"2026-05-24T02:07:18.178Z","message":{"role":"assistant","content":[],"api":"openai-responses","provider":"openai_proxy","model":"gpt-5.4-mini","usage":{"input":0,"output":0,"cacheRead":0,"cacheWrite":0,"totalTokens":0,"cost":{"input":0,"output":0,"cacheRead":0,"cacheWrite":0,"total":0}},"stopReason":"error","timestamp":1779588434585,"errorMessage":"503 Service temporarily unavailable"}}
{"type":"message","id":"67ead47b","parentId":"d92d6fec","timestamp":"2026-05-24T02:07:23.201Z","message":{"role":"assistant","content":[],"api":"openai-responses","provider":"openai_proxy","model":"gpt-5.4-mini","usage":{"input":0,"output":0,"cacheRead":0,"cacheWrite":0,"totalTokens":0,"cost":{"input":0,"output":0,"cacheRead":0,"cacheWrite":0,"total":0}},"stopReason":"error","timestamp":1779588440189,"errorMessage":"503 Service temporarily unavailable"}}
{"type":"message","id":"7adf5274","parentId":"67ead47b","timestamp":"2026-05-24T02:07:30.602Z","message":{"role":"assistant","content":[],"api":"openai-responses","provider":"openai_proxy","model":"gpt-5.4-mini","usage":{"input":0,"output":0,"cacheRead":0,"cacheWrite":0,"totalTokens":0,"cost":{"input":0,"output":0,"cacheRead":0,"cacheWrite":0,"total":0}},"stopReason":"error","timestamp":1779588447216,"errorMessage":"503 Service temporarily unavailable"}}
{"type":"message","id":"e58ccadf","parentId":"7adf5274","timestamp":"2026-05-24T02:07:42.221Z","message":{"role":"assistant","content":[],"api":"openai-responses","provider":"openai_proxy","model":"gpt-5.4-mini","usage":{"input":0,"output":0,"cacheRead":0,"cacheWrite":0,"totalTokens":0,"cost":{"input":0,"output":0,"cacheRead":0,"cacheWrite":0,"total":0}},"stopReason":"error","timestamp":1779588458616,"errorMessage":"503 Service temporarily unavailable"}}
{"type":"custom","customType":"model-snapshot","data":{"timestamp":1779588462261,"provider":"hai","modelApi":"openai-completions","modelId":"MiniMax-M2.5"},"id":"b7de3f6f","parentId":"e58ccadf","timestamp":"2026-05-24T02:07:42.261Z"}
{"type":"message","id":"af8bbcfd","parentId":"b7de3f6f","timestamp":"2026-05-24T02:07:42.273Z","message":{"role":"user","content":[{"type":"text","text":"System: [2026-05-24 10:07:13 GMT+8] Slack DM from 李力: 有没有类似于hermes的新的这种项目出现\n\nConversation info (untrusted metadata):\n```json\n{\n  \"message_id\": \"1779588432.621559\",\n  \"reply_to_id\": \"1779588432.621559\",\n  \"sender_id\": \"U0A8ZCK7RJM\",\n  \"sender\": \"李力\",\n  \"timestamp\": \"Sun 2026-05-24 10:07 GMT+8\"\n}\n```\n\nSender (untrusted metadata):\n```json\n{\n  \"label\": \"李力 (U0A8ZCK7RJM)\",\n  \"id\": \"U0A8ZCK7RJM\",\n  \"name\": \"李力\"\n}\n```\n\n有没有类似于hermes的新的这种项目出现"}],"timestamp":1779588462267}}
{"type":"message","id":"34ae5359","parentId":"af8bbcfd","timestamp":"2026-05-24T02:07:50.043Z","message":{"role":"assistant","content":[{"type":"thinking","thinking":"The user (李力) is asking about new projects similar to Hermes. I don't have context about what \"hermes\" refers to specifically - it could be Hermes (Meta's JavaScript engine), Hermes (the AI framework?), or something else. Let me search for recent projects similar to Hermes to give a good answer.\n\nActually, looking at the message, it says \"类似于hermes的新的这种项目出现\" - asking if there are new projects similar to Hermes. This seems like a tech question. Without more context about which Hermes project they're referring to, let me search for recent alternatives/successors to Hermes (the JS engine by Meta, or possibly another Hermes project).\n\n\nI'll do a web search to find what's currently trending as alternatives or next-generation projects in this space."},{"type":"toolCall","id":"call_d49119922db14356bef990f3","name":"web_search","arguments":{"query":"new projects similar to Hermes AI agent framework 2025 2026 alternatives","count":8,"freshness":"year"}}],"api":"openai-completions","provider":"hai","model":"MiniMax-M2.5","usage":{"input":24433,"output":218,"cacheRead":130,"cacheWrite":0,"totalTokens":24781,"cost":{"input":0,"output":0,"cacheRead":0,"cacheWrite":0,"total":0}},"stopReason":"toolUse","timestamp":1779588462272}}
{"type":"message","id":"a9bfe7a9","parentId":"34ae5359","timestamp":"2026-05-24T02:08:05.456Z","message":{"role":"toolResult","toolCallId":"call_d49119922db14356bef990f3","toolName":"web_search","content":[{"type":"text","text":"{\n  \"query\": \"new projects similar to Hermes AI agent framework 2025 2026 alternatives\",\n  \"provider\": \"perplexity\",\n  \"model\": \"perplexity/sonar-pro\",\n  \"tookMs\": 15385,\n  \"externalContent\": {\n    \"untrusted\": true,\n    \"source\": \"web_search\",\n    \"provider\": \"perplexity\",\n    \"wrapped\": true\n  },\n  \"content\": \"\\n<<<EXTERNAL_UNTRUSTED_CONTENT id=\\\"711cf3693ff42a13\\\">>>\\nSource: Web Search\\n---\\nHere are the main 2025–2026 projects that are closest in spirit to Hermes Agent (autonomous / persistent / tool-using agents), plus how they differ and when you’d pick each.\\n\\n---\\n\\n## 1. OpenClaw / OmniClaw\\n\\n**What it is:**  \\nOpen-source, self-hosted browser agent framework focused on web navigation and automation (multi-step browsing, form-filling, sign-ins, etc.).\\n\\n**Why it’s similar to Hermes Agent**\\n- Self‑hosted, agentic framework, can run on cheap VPS.\\n- Focus on autonomy and multi-step task execution.\\n- Uses LLMs via API; you bring your own model.\\n\\n**Key differences**\\n- **Browser-first**: built around controlling a headless browser (or system browser), especially for web tasks.\\n- **CLI-centric** and “channels everywhere” (terminal, TUI, messaging, webhooks).\\n- Ecosystem reach vs “learning loop”: OpenClaw emphasizes integrating with many surfaces; Hermes emphasizes persistent memory and auto-learned skills.\\n\\n**Good use cases**\\n- Agents that need to reliably use real websites (dashboards, SaaS tools, scraping + interaction).\\n- Devs who want a simple, scriptable CLI agent with wide OS support.\\n\\n---\\n\\n## 2. Vellum (Local Personal Agent)\\n\\n**What it is:**  \\nLocal personal AI assistant that runs on your device, with its own identity, persistent memory, and local credentials handling.\\n\\n**Why it’s similar**\\n- “Personal AI that grows with you” angle, like Hermes’ “agent that grows with you.”\\n- Long-term memory, identity, and evolving behavior over time.\\n\\n**Differences**\\n- **Device-first** rather than server/cluster-first.\\n- Strong focus on **privacy** and keeping credentials off remote models.\\n- More “personal sidekick” than “multi-tenant server agent framework.”\\n\\n**Good use cases**\\n- Personal desktop assistant with long-term relationship and memory.\\n- Users who want privacy and local control rather than a hosted agent stack.\\n\\n---\\n\\n## 3. Claude Cowork (Anthropic)\\n\\n**What it is:**  \\nAnthropic’s “Cowork” desktop agent: Claude can see your screen and control mouse/keyboard, doing actual desktop workflows.\\n\\n**Similarities**\\n- Autonomous execution of complex tasks.\\n- Tool use and multi-step planning.\\n\\n**Differences**\\n- **Managed, closed-source product**, not a framework.\\n- Desktop / screen-control orientation, not a programmable framework like Hermes Agent.\\n- Integrated tightly with Claude models and Anthropic’s ecosystem.\\n\\n**Good use cases**\\n- Knowledge work on a desktop: inbox triage, doc editing, spreadsheet tooling, etc.\\n- Users who want hands-off work completion without building infra.\\n\\n---\\n\\n## 4. Manus\\n\\n**What it is:**  \\nCloud-first, long-horizon autonomous agent; you hand it a complex task and it works for long periods. Launched 2025, acquired by Meta.\\n\\n**Similarities**\\n- Focus on autonomy and long-running tasks.\\n- Persistent “project-level” memory.\\n\\n**Differences**\\n- Hosted SaaS with proprietary stack.\\n- Geared towards multi-hour, high-level deliverables rather than being an OSS framework to integrate into your own stack.\\n\\n**Good use cases**\\n- Delegating substantial, long-running projects (reports, research, multi-day workflows) with minimal setup.\\n- Teams who prefer a managed product to DIY infra.\\n\\n---\\n\\n## 5. OpenAI Operator\\n\\n**What it is:**  \\nOpenAI’s autonomous browser agent (“research preview” launched Feb 2025) that completes web tasks for you.\\n\\n**Similarities**\\n- Browser agent, similar autonomy paradigm.\\n- Tool-using, multi-step reasoning.\\n\\n**Differences**\\n- Closed-source, tied to OpenAI ecosystem.\\n- Less of a “framework” and more a feature inside ChatGPT.\\n\\n**Good use cases**\\n- People already using ChatGPT Pro who want automated web task completion with zero setup.\\n- Not ideal if you want deep customization or self-hosting.\\n\\n---\\n\\n## 6. Sai (Simular)\\n\\n**What it is:**  \\nDesktop AI agent by Simular (ex‑DeepMind team). Focus on on-device agent with autonomy and persistent behavior.\\n\\n**Similarities**\\n- Desktop personal agent with memory, autonomy, and “companion” feel.\\n- Evolving behavior over time.\\n\\n**Differences**\\n- Desktop GUI/agent vs server framework.\\n- Emphasizes safety and guardrails out of the box.\\n\\n**Good use cases**\\n- Personal desktop agents with a strong safety / research pedigree.\\n- Users who want an out‑of‑the‑box agent rather than integrating SDKs.\\n\\n---\\n\\n## 7. Superagent\\n\\n**What it is:**  \\nOpen-source agent framework for building custom agents and automation, often used by internal teams.\\n\\n**Similarities**\\n- Framework for building custom tools + LLM agents.\\n- Designed for server deployment and integration into existing systems.\\n\\n**Differences**\\n- Less focused (today) on a baked-in self-learning loop; more of a general-purpose “API + runtime” for agents.\\n- Often used in B2B/internal automation contexts.\\n\\n**Good use cases**\\n- Teams wanting OSS plus a cleaner, production-ready agent API and dashboard.\\n- Internal automation across many APIs/services.\\n\\n---\\n\\n## 8. AutoGen, LangGraph, CrewAI, Semantic Kernel, MetaGPT, etc.\\n\\nThese are older than 2025 but still evolving and relevant:\\n\\n- **AutoGen (Microsoft Research)**  \\n  Multi-agent conversation framework; great for research and multi-role setups.\\n- **LangGraph**  \\n  Graph-based workflows to structure multi-step agent tasks.\\n- **CrewAI**  \\n  Multi-agent orchestration; popular for “team of agents” patterns.\\n- **Semantic Kernel**  \\n  .NET / Azure-focused SDK for building agents with skills, planners, and memory.\\n- **MetaGPT**  \\n  Multi-agent framework modeling a “software company” (PM, dev, QA) to build software.\\n\\n**Why they’re alternatives**\\n- All provide tools to orchestrate tool-calling LLMs, memory, and multi-step workflows.\\n- You can often replicate Hermes-like behavior, though you may manage state/memory more manually.\\n\\n**When to pick them**\\n- You want structured, multi-agent systems or are already in an ecosystem (.NET/Azure, Python research, etc.).\\n- You’re comfortable owning more of the memory/state logic yourself.\\n\\n---\\n\\n## 9. Pydantic AI, Mastra, LlamaIndex + Filesystem Tools\\n\\nNot pure “Hermes Agent equivalents,” but worth mentioning:\\n\\n- **Pydantic AI** (Python) & **Mastra** (TypeScript)  \\n  Tool + agent definitions with strong typing and good dev ergonomics.\\n- **LlamaIndex** (esp. filesystem tools)  \\n  Modern RAG + agent patterns that can behave like task-focused agents.\\n\\n**Good for**\\n- Dev teams who care about type safety and want fine control over orchestration, not a batteries-included agent runtime.\\n\\n---\\n\\n## 10. Dot (GAgentDot/dot)\\n\\n**What it is:**  \\nTiny desktop companion powered by Claude, with self-RL, self-rewrite, container isolation, multiple “moods.”\\n\\n**Similarities**\\n- Self-improvement loop: the agent rewrites itself and learns over time.\\n- Persistent, always-on companion style.\\n\\n**Differences**\\n- Very “toy/pet” aesthetic, but architecturally serious (RL, sandboxing).\\n- Desktop companion rather than generic backend framework.\\n\\n**Good use cases**\\n- Experimenting with self-improving, persistent agents running locally.\\n- Using Claude-powered autonomy in a contained environment.\\n\\n---\\n\\n## How to choose among them (quick mapping)\\n\\n- **Want a self-hosted agent framework with persistent memory and learning loop:**  \\n  Hermes Agent, OpenClaw, Superagent.\\n- **Want a personal desktop agent that grows with you:**  \\n  Vellum, Sai, Dot, Claude Cowork.\\n- **Want long-horizon managed autonomy (cloud):**  \\n  Manus, Perplexity Computer, OpenAI Operator.\\n- **Want a composable dev framework for custom systems:**  \\n  AutoGen, LangGraph, CrewAI, Semantic Kernel, Pydantic AI, Mastra, LlamaIndex.\\n\\nIf you tell me your stack (Python/TS, self‑hosted vs managed, web vs desktop vs backend) and priority (privacy, cost, control, ease of use), I can narrow this down to 2–3 concrete recommendations with links and starter patterns.\\n<<<END_EXTERNAL_UNTRUSTED_CONTENT id=\\\"711cf3693ff42a13\\\">>>\",\n  \"citations\": [\n    \"https://espressio.ai/blog/hermes-agent-guide-2026/\",\n    \"https://www.linkedin.com/posts/aaabulkhair_ai-llms-aiagents-activity-7447243609256484864-1fgG\",\n    \"https://www.simular.ai/alternatives/openclaw-alternatives\",\n    \"https://www.vellum.ai/blog/best-hermes-agent-alternatives\",\n    \"https://www.reddit.com/r/AI_Agents/comments/1smyloy/what_frameworks_are_currently_best_for_building/\",\n    \"https://thenewstack.io/persistent-ai-agents-compared/\",\n    \"https://www.facebook.com/groups/openclawgroup/posts/1878561446183754/\",\n    \"https://www.youtube.com/watch?v=gmaHRwijOXs&vl=en\",\n    \"https://www.firecrawl.dev/blog/best-open-source-agent-frameworks\",\n    \"https://composio.dev/content/best-ai-agent-builders-and-integrations\"\n  ]\n}"}],"details":{"query":"new projects similar to Hermes AI agent framework 2025 2026 alternatives","provider":"perplexity","model":"perplexity/sonar-pro","tookMs":15385,"externalContent":{"untrusted":true,"source":"web_search","provider":"perplexity","wrapped":true},"content":"\n<<<EXTERNAL_UNTRUSTED_CONTENT id=\"711cf3693ff42a13\">>>\nSource: Web Search\n---\nHere are the main 2025–2026 projects that are closest in spirit to Hermes Agent (autonomous / persistent / tool-using agents), plus how they differ and when you’d pick each.\n\n---\n\n## 1. OpenClaw / OmniClaw\n\n**What it is:**  \nOpen-source, self-hosted browser agent framework focused on web navigation and automation (multi-step browsing, form-filling, sign-ins, etc.).\n\n**Why it’s similar to Hermes Agent**\n- Self‑hosted, agentic framework, can run on cheap VPS.\n- Focus on autonomy and multi-step task execution.\n- Uses LLMs via API; you bring your own model.\n\n**Key differences**\n- **Browser-first**: built around controlling a headless browser (or system browser), especially for web tasks.\n- **CLI-centric** and “channels everywhere” (terminal, TUI, messaging, webhooks).\n- Ecosystem reach vs “learning loop”: OpenClaw emphasizes integrating with many surfaces; Hermes emphasizes persistent memory and auto-learned skills.\n\n**Good use cases**\n- Agents that need to reliably use real websites (dashboards, SaaS tools, scraping + interaction).\n- Devs who want a simple, scriptable CLI agent with wide OS support.\n\n---\n\n## 2. Vellum (Local Personal Agent)\n\n**What it is:**  \nLocal personal AI assistant that runs on your device, with its own identity, persistent memory, and local credentials handling.\n\n**Why it’s similar**\n- “Personal AI that grows with you” angle, like Hermes’ “agent that grows with you.”\n- Long-term memory, identity, and evolving behavior over time.\n\n**Differences**\n- **Device-first** rather than server/cluster-first.\n- Strong focus on **privacy** and keeping credentials off remote models.\n- More “personal sidekick” than “multi-tenant server agent framework.”\n\n**Good use cases**\n- Personal desktop assistant with long-term relationship and memory.\n- Users who want privacy and local control rather than a hosted agent stack.\n\n---\n\n## 3. Claude Cowork (Anthropic)\n\n**What it is:**  \nAnthropic’s “Cowork” desktop agent: Claude can see your screen and control mouse/keyboard, doing actual desktop workflows.\n\n**Similarities**\n- Autonomous execution of complex tasks.\n- Tool use and multi-step planning.\n\n**Differences**\n- **Managed, closed-source product**, not a framework.\n- Desktop / screen-control orientation, not a programmable framework like Hermes Agent.\n- Integrated tightly with Claude models and Anthropic’s ecosystem.\n\n**Good use cases**\n- Knowledge work on a desktop: inbox triage, doc editing, spreadsheet tooling, etc.\n- Users who want hands-off work completion without building infra.\n\n---\n\n## 4. Manus\n\n**What it is:**  \nCloud-first, long-horizon autonomous agent; you hand it a complex task and it works for long periods. Launched 2025, acquired by Meta.\n\n**Similarities**\n- Focus on autonomy and long-running tasks.\n- Persistent “project-level” memory.\n\n**Differences**\n- Hosted SaaS with proprietary stack.\n- Geared towards multi-hour, high-level deliverables rather than being an OSS framework to integrate into your own stack.\n\n**Good use cases**\n- Delegating substantial, long-running projects (reports, research, multi-day workflows) with minimal setup.\n- Teams who prefer a managed product to DIY infra.\n\n---\n\n## 5. OpenAI Operator\n\n**What it is:**  \nOpenAI’s autonomous browser agent (“research preview” launched Feb 2025) that completes web tasks for you.\n\n**Similarities**\n- Browser agent, similar autonomy paradigm.\n- Tool-using, multi-step reasoning.\n\n**Differences**\n- Closed-source, tied to OpenAI ecosystem.\n- Less of a “framework” and more a feature inside ChatGPT.\n\n**Good use cases**\n- People already using ChatGPT Pro who want automated web task completion with zero setup.\n- Not ideal if you want deep customization or self-hosting.\n\n---\n\n## 6. Sai (Simular)\n\n**What it is:**  \nDesktop AI agent by Simular (ex‑DeepMind team). Focus on on-device agent with autonomy and persistent behavior.\n\n**Similarities**\n- Desktop personal agent with memory, autonomy, and “companion” feel.\n- Evolving behavior over time.\n\n**Differences**\n- Desktop GUI/agent vs server framework.\n- Emphasizes safety and guardrails out of the box.\n\n**Good use cases**\n- Personal desktop agents with a strong safety / research pedigree.\n- Users who want an out‑of‑the‑box agent rather than integrating SDKs.\n\n---\n\n## 7. Superagent\n\n**What it is:**  \nOpen-source agent framework for building custom agents and automation, often used by internal teams.\n\n**Similarities**\n- Framework for building custom tools + LLM agents.\n- Designed for server deployment and integration into existing systems.\n\n**Differences**\n- Less focused (today) on a baked-in self-learning loop; more of a general-purpose “API + runtime” for agents.\n- Often used in B2B/internal automation contexts.\n\n**Good use cases**\n- Teams wanting OSS plus a cleaner, production-ready agent API and dashboard.\n- Internal automation across many APIs/services.\n\n---\n\n## 8. AutoGen, LangGraph, CrewAI, Semantic Kernel, MetaGPT, etc.\n\nThese are older than 2025 but still evolving and relevant:\n\n- **AutoGen (Microsoft Research)**  \n  Multi-agent conversation framework; great for research and multi-role setups.\n- **LangGraph**  \n  Graph-based workflows to structure multi-step agent tasks.\n- **CrewAI**  \n  Multi-agent orchestration; popular for “team of agents” patterns.\n- **Semantic Kernel**  \n  .NET / Azure-focused SDK for building agents with skills, planners, and memory.\n- **MetaGPT**  \n  Multi-agent framework modeling a “software company” (PM, dev, QA) to build software.\n\n**Why they’re alternatives**\n- All provide tools to orchestrate tool-calling LLMs, memory, and multi-step workflows.\n- You can often replicate Hermes-like behavior, though you may manage state/memory more manually.\n\n**When to pick them**\n- You want structured, multi-agent systems or are already in an ecosystem (.NET/Azure, Python research, etc.).\n- You’re comfortable owning more of the memory/state logic yourself.\n\n---\n\n## 9. Pydantic AI, Mastra, LlamaIndex + Filesystem Tools\n\nNot pure “Hermes Agent equivalents,” but worth mentioning:\n\n- **Pydantic AI** (Python) & **Mastra** (TypeScript)  \n  Tool + agent definitions with strong typing and good dev ergonomics.\n- **LlamaIndex** (esp. filesystem tools)  \n  Modern RAG + agent patterns that can behave like task-focused agents.\n\n**Good for**\n- Dev teams who care about type safety and want fine control over orchestration, not a batteries-included agent runtime.\n\n---\n\n## 10. Dot (GAgentDot/dot)\n\n**What it is:**  \nTiny desktop companion powered by Claude, with self-RL, self-rewrite, container isolation, multiple “moods.”\n\n**Similarities**\n- Self-improvement loop: the agent rewrites itself and learns over time.\n- Persistent, always-on companion style.\n\n**Differences**\n- Very “toy/pet” aesthetic, but architecturally serious (RL, sandboxing).\n- Desktop companion rather than generic backend framework.\n\n**Good use cases**\n- Experimenting with self-improving, persistent agents running locally.\n- Using Claude-powered autonomy in a contained environment.\n\n---\n\n## How to choose among them (quick mapping)\n\n- **Want a self-hosted agent framework with persistent memory and learning loop:**  \n  Hermes Agent, OpenClaw, Superagent.\n- **Want a personal desktop agent that grows with you:**  \n  Vellum, Sai, Dot, Claude Cowork.\n- **Want long-horizon managed autonomy (cloud):**  \n  Manus, Perplexity Computer, OpenAI Operator.\n- **Want a composable dev framework for custom systems:**  \n  AutoGen, LangGraph, CrewAI, Semantic Kernel, Pydantic AI, Mastra, LlamaIndex.\n\nIf you tell me your stack (Python/TS, self‑hosted vs managed, web vs desktop vs backend) and priority (privacy, cost, control, ease of use), I can narrow this down to 2–3 concrete recommendations with links and starter patterns.\n<<<END_EXTERNAL_UNTRUSTED_CONTENT id=\"711cf3693ff42a13\">>>","citations":["https://espressio.ai/blog/hermes-agent-guide-2026/","https://www.linkedin.com/posts/aaabulkhair_ai-llms-aiagents-activity-7447243609256484864-1fgG","https://www.simular.ai/alternatives/openclaw-alternatives","https://www.vellum.ai/blog/best-hermes-agent-alternatives","https://www.reddit.com/r/AI_Agents/comments/1smyloy/what_frameworks_are_currently_best_for_building/","https://thenewstack.io/persistent-ai-agents-compared/","https://www.facebook.com/groups/openclawgroup/posts/1878561446183754/","https://www.youtube.com/watch?v=gmaHRwijOXs&vl=en","https://www.firecrawl.dev/blog/best-open-source-agent-frameworks","https://composio.dev/content/best-ai-agent-builders-and-integrations"]},"isError":false,"timestamp":1779588485440}}
{"type":"message","id":"ee28e2fe","parentId":"a9bfe7a9","timestamp":"2026-05-24T02:08:18.548Z","message":{"role":"assistant","content":[{"type":"thinking","thinking":"李力问的是有没有类似 Hermes 的新项目。搜索结果给出了很多信息，我来整理一下给李力的回答。"},{"type":"text","text":"李力，你问的是有没有类似 Hermes 的新项目，我刚搜了一下，近两年确实冒出不少类似的框架，给你归个类：\n\n## 🏆 最像 Hermes 的新一代项目\n\n**OpenClaw** — 跟 Hermes 最像的，开源、自托管、支持持久记忆、多步任务执行，CLI 为主但能对接各种消息渠道（Slack、飞书、QQ 等）。我自己就在用这个跑在 VPS 上。\n\n**Manus** — 2025 年上线，专注长时自主任务，云端运行，丢一个复杂任务进去它能跑好几个小时给你出结果，已被 Meta 收购。\n\n**Superagent** — 开源 agent 框架，设计思路和 Hermes 接近，支持自定义工具和持久记忆，适合团队内建 agent 平台。\n\n## 🧩 开发者友好的编排框架（偏底层）\n\n| 框架 | 特点 |\n|------|------|\n| **AutoGen**（微软） | 多 agent 对话协作 |\n| **CrewAI** | 多 agent 团队编排，很流行 |\n| **LangGraph** | 图结构任务流 |\n| **Mastra**（TS） | 类型安全，DevEx 好 |\n| **Pydantic AI**（Python） | 强类型 + 工具调用 |\n\n## 🖥️ 桌面端个人 agent\n\n- **Claude Cowork**（Anthropic）— 看屏幕、控制鼠标键盘帮你干活\n- **Sai**（Simular）— 桌面 agent，专注安全和持久行为\n- **Dot** — 小型本地 companion，self-RL 自进化\n\n## 怎么选\n\n- **想自托管 + 跟 Hermes 体验最接近** → OpenClaw\n- **想要 managed 不用自己维护** → Manus\n- **团队内建 agent 平台** → Superagent 或 CrewAI\n- **自己写微调控制** → AutoGen / LangGraph / Mastra\n\n你是想自己搭一套，还是找现成的产品用？说说你的场景我帮你细筛。"}],"api":"openai-completions","provider":"hai","model":"MiniMax-M2.5","usage":{"input":2447,"output":440,"cacheRead":24560,"cacheWrite":0,"totalTokens":27447,"cost":{"input":0,"output":0,"cacheRead":0,"cacheWrite":0,"total":0}},"stopReason":"stop","timestamp":1779588485455}}