OpenClaw
An AI agent tool built around a skill catalog called ClawHub. Released in November 2025, it features over 137 version iterations and more than 5,400 skills, with support for tool invocation, memory management, and gateway functionality.
Core Facts
Limited by the computing power and memory of consumer-grade devices, the scale of models that can be run locally is typically far smaller than cloud-based flagship models, with gaps in complex reasoning, long-context processing, and other tasks.
A Skill is a modular encapsulation that grants an agent specialized capabilities, containing complete workflows, judgment logic, and contextual knowledge, and can combine multiple tools to accomplish complex goals.
OpenClaw is an open-source AI Agent framework with tool-calling, memory management, and autonomous evolution capabilities.
Timeline (last 90 days)
OpenClaw's doctor detects problems in three layers (configuration anomalies, state problems, service health), repairs them in two tiers (safe repairs are executed automatically, risky ones require user confirmation), and hands problems that deterministic rules cannot cover to the LLM to analyze logs, infer causes, and generate repair plans.
OpenClaw uses MCP through the mcporter bridge, so every poisoned skill is a potential MCP attack vector.
In the ClawHavoc attack, more than 800 malicious skills were uploaded to the OpenClaw marketplace, triggering 1,142 security advisories and exposing more than 135,000 instances.
Microsoft lists GitHub Copilot, OpenAI Codex, Replit, OpenClaw, LM Studio, and Unsloth AI as currently supporting MXC, with Claude Code and others announced as coming soon; Entra-based agent identity is described as "coming soon" and has not yet shipped.
Mem0's OpenClaw integration describes a 'structured protocol with an importance gate and domain coverage'; in the triage protocol, the Agent performs four gate judgments on each candidate fact, among which Gate 2 (NOVELTY) checks whether the fact has already been recalled.
Atomic Agent's supported import sources include Claude Code, Codex, Hermes, OpenClaw, Pi and Oh-My-Pi, covering dimensions such as skills, memory, MCP servers and session records; among them, Hermes's and OpenClaw's memory data cannot be migrated, because the two have different storage formats and no exchange standard.
NIBVOK AI Security v0.1.2 is a runtime enforcement layer for OpenClaw agents.
The OpenClaw gateway runs on port 8080.
Integration can be achieved by pointing OpenClaw to http://localhost:11434/v1 and setting the model to qwen3.5:35b, with quantization, GPU offloading, and context management handled automatically by Ollama.
Some users accidentally spent tens of thousands or even over a hundred thousand yuan while using OpenClaw.
40 more timeline events
All Facts (20)
Limited by the computing power and memory of consumer-grade devices, the scale of models that can be run locally is typically far smaller than cloud-based flagship models, with gaps in complex reasoning, long-context processing, and other tasks.
90%VerifiedA Skill is a modular encapsulation that grants an agent specialized capabilities, containing complete workflows, judgment logic, and contextual knowledge, and can combine multiple tools to accomplish complex goals.
90%VerifiedOpenClaw is an open-source AI Agent framework with tool-calling, memory management, and autonomous evolution capabilities.
80%VerifiedOpenClaw's architecture is: a person connects to OpenClaw (local computer) via communication software (WhatsApp, Telegram, etc.), which then connects to a language model (cloud/local); OpenClaw itself is an interface program between the person and the language model and does not possess independent intelligence.
65%UnverifiedOpenClaw支持SubAgent机制，语言模型可以要求生成子Agent来处理子任务，但程序端禁止子Agent再使用繁殖工具以防止无限递归
95%Unverifiedopenclaw gateway status 命令用于检查OpenClaw Gateway的运行状态，RPC状态显示OK表示服务正常运行
95%Unverifiedopenclaw dashboard 命令用于打开OpenClaw自带的Web UI界面，但需要Gateway启动后才能正常加载
95%UnverifiedOpenClaw使用Compaction机制解决Context Window不够用的问题，将较旧的对话历史发给语言模型做摘要，用摘要替换原文，该过程可递归进行
90%UnverifiedOpenClaw的心跳机制每隔固定时间（如30分钟）自动触发语言模型，让它读取habit.md中的待办事项并执行，配合CronJob排程系统实现定时任务
90%UnverifiedOpenClaw的记忆检索本质上是RAG（检索增强生成），将记忆文件切成Chunk，通过字面匹配和语义Embedding的加权相似度排序，取Top-K结果返回给语言模型
90%UnverifiedOpenClaw中最强大也最危险的工具是Execute，可以执行任何Shell命令，存在语言模型执行破坏性命令（如rm -rf）的风险
90%Unverifiednotebooklm-py可作为Claude Code、Codex、OpenClaw等AI智能体的工具插件
90%UnverifiedClaude Code和OpenClaw都可以将飞书CLI作为外部工具调用，从而获得操作飞书的能力
90%UnverifiedARIS采用纯Markdown技能系统，不依赖任何特定框架，能与Claude Code、Codex、OpenClaw或任意LLM代理配合
90%UnverifiedMoltbot这个名字只存活了三天（2026-01-27至2026-01-30）就被替换为最终的OpenClaw
90%UnverifiedOpenClaw项目经历了六次命名：Warelay（2025-11-24）→ CLAWDIS（2025-12-03）→ CLAWDBOT（2026-01-04）→ Clawdbot（2026-01-10）→ Moltbot（2026-01-27）→ OpenClaw（2026-01-30）
90%UnverifiedOpenClaw项目最初名为Warelay，于2025年11月24日首次提交，定位为基于Twilio的WhatsApp中继CLI工具
90%Unverifiedsoul.md和user.md配置文件写得过于冗长会在每轮对话开始时预消耗大量token，导致模型能处理的实际对话内容减少
90%UnverifiedOpenClaw的Dashboard只是前端展示层，真正的业务逻辑全部由Gateway驱动
90%UnverifiedOpenClaw需要重启的三种主要场景是：修改了配置文件、添加了新的工具插件（Tool而非skill）、龙虾不响应
90%