开源AI工作流 是 AI Skill Hub 本期精选Agent工作流之一。综合评分 7.5 分，整体质量较高。我们推荐使用将其纳入你的 AI 工具库，帮助提升工作效率。
An installable AI operating system — persistent agents with living brains，提供高质量的AI工作流解决方案
开源AI工作流 是一套完整的 AI Agent 自动化工作流方案。通过可视化的节点编排，将复杂的多步骤任务拆解为清晰的自动化流程，实现全程无人值守的智能处理。支持与数百种外部服务和 API 无缝集成，适合构建数据处理管线、业务自动化和 AI 辅助决策系统。
An installable AI operating system — persistent agents with living brains，提供高质量的AI工作流解决方案
开源AI工作流 是一套完整的 AI Agent 自动化工作流方案。通过可视化的节点编排，将复杂的多步骤任务拆解为清晰的自动化流程，实现全程无人值守的智能处理。支持与数百种外部服务和 API 无缝集成，适合构建数据处理管线、业务自动化和 AI 辅助决策系统。
# 方式一：npm 全局安装 npm install -g home23 # 方式二：npx 直接运行（无需安装） npx home23 --help # 方式三：项目依赖安装 npm install home23 # 方式四：从源码运行 git clone https://github.com/notforyou23/home23 cd home23 npm install npm start
# 命令行使用
home23 --help
# 基本用法
home23 [options] <input>
# Node.js 代码中使用
const home23 = require('home23');
const result = await home23.run(options);
console.log(result);
# home23 配置说明 # 查看配置选项 home23 --config-example > config.yml # 常见配置项 # output_dir: ./output # log_level: info # workers: 4 # 环境变量（覆盖配置文件） export HOME23_CONFIG="/path/to/config.yml"
An installable AI operating system — persistent agents with living brains.
Home23 is not another chatbot framework. It is a complete AI operating system that runs on your machine, with agents that think autonomously, grow a persistent brain over time, dream during idle periods, and are reachable through Telegram, Discord, a web dashboard, an AI IDE, a research engine, and a mobile-optimized chat page you can Add to Home Screen on iOS.
Documentation status: refreshed 2026-05-16 against the running Home23 app. The release metadata, README, changelog, manifest, and dashboard welcome/version banner now agree on v0.6.0. Older step-by-step design files in docs/design/ are historical build records; use this README plus CLAUDE.md, AGENTS.md, CHANGELOG.md, and docs/MANIFEST.md for the current public repo shape.
npm install -g pm2)git clone https://github.com/notforyou23/home23.git
cd home23
node cli/home23.js initInit checks prerequisites, installs all dependencies (including MarkItDown for document ingestion), sets up encryption keys and the OAuth database, builds the TypeScript harness, and seeds all configuration. No API keys needed — provider setup happens in the web dashboard.
Everything after init happens in the browser. Start the system:
node cli/home23.js startThen open http://localhost:5002/home23 in your browser.
First-run flow — a guided onboarding wizard walks you through setup: 1. Welcome screen — click "Get Started" 2. Providers — sign in with OAuth (recommended for Anthropic and OpenAI Codex) or enter API keys. At least one provider required to continue. 3. Create Agent — name, owner, model selection (shows only models from your configured providers), optional Telegram token 4. Launch — starts the agent and takes you to the dashboard home screen 5. You're live — real-time thoughts, chat, intelligence synthesis, and full system access
The web dashboard is the primary interface for everything — provider configuration, agent creation, model selection, feeder settings, and day-to-day use. The CLI handles init, start/stop, and updates.
What you see: - Dashboard: http://localhost:5002/home23 — OS home screen - Settings: http://localhost:5002/home23/settings — all configuration - Chat: http://localhost:5002/home23/chat — standalone chat page (mobile-first; bookmark on your phone or use Safari → Share → Add to Home Screen to launch as a full-screen PWA) - Evobrew IDE: http://localhost:3415 — AI code editor with brain access - COSMO Research: http://localhost:43210 — research engine UI
COSMO Query uses normal direct Query for small completed brains and Progressive Graph Search for large-graph coverage. As of Patch 20, small PGS requests fall back to direct Query and one-partition graphs skip fake cross-partition synthesis; vendored COSMO changes are tracked in docs/design/COSMO23-VENDORED-PATCHES.md.
| File | Purpose |
|---|---|
config/home.yaml | Provider URLs, model aliases, embedding config, chat defaults |
config/secrets.yaml | API keys and bot tokens (managed by dashboard Settings, gitignored) |
instances/<name>/config.yaml | Per-agent: ports, owner, channels (Telegram/Discord/iMessage/webhooks), chat model, scheduler, modelAssignments overrides |
configs/base-engine.yaml | Cognitive loop timing, feeder block, default modelAssignments (shared across agents) |
configs/action-allowlist.yaml | Autonomous action governance — enabled actions, rate limits, dry-run flags, integrations |
The pulse LLM brief includes a --- LIVE PROBLEMS (verified just now — ground truth) --- block. The system prompt hard-blocks the agent from asserting anything is broken unless it appears in that block. Status changes only (newly opened, chronic, resolved-just-now). Stable-open problems are NOT restated. This eliminates the stale-assertion loop.
开源AI工作流提供了一个高质量的AI工作流解决方案，适合于开发和管理AI工作流，值得一试
AI Skill Hub 为第三方内容聚合平台，本页面信息基于公开数据整理，不对工具功能和质量作任何法律背书。
建议在沙箱或测试环境中充分验证后，再部署至生产环境，并做好必要的安全评估。
✅ MIT 协议 — 最宽松的开源协议之一，可自由商用、修改、分发，仅需保留版权声明。
经综合评估，开源AI工作流 在Agent工作流赛道中表现稳健，质量良好。如果你已有明确的使用需求，可以直接上手体验；如果还在评估阶段，建议对比同类工具后再做决策。
| 原始名称 | home23 |
| Topics | workflowjavascript |
| GitHub | https://github.com/notforyou23/home23 |
| License | MIT |
| 语言 | JavaScript |
收录时间：2026-05-23 · 更新时间：2026-05-23 · License：MIT · AI Skill Hub 不对第三方内容的准确性作法律背书。
选择 Agent 类型，复制安装指令后粘贴到对应客户端