🔐 TrustClaw：Composio 出品 575★ 自托管 AI Agent，一秒部署永不掉线，你睡觉它干活
TrustClaw 是 Composio（28k★ 的 AI 工具集成平台）刚开源的个人 AI 助手。跟那些要你配半天 API Key、自己管 prompt 的工具不同，TrustClaw 主打一个「部署即用」：
npx @composio/trustclaw deploy
一行命令，两分钟，你就有了一个 24/7 在线、会聊 Telegram、能调 1000+ 工具（Gmail/GitHub/Slack/Notion/Stripe…）、有长期记忆、还能定时任务自动执行的全职 AI 助手。
不用 Anthropic 或 OpenAI 的 API Key——LLM 调用走 Vercel AI Gateway，开箱即用。
它解决了什么？
本地跑 AI Agent 最大的痛点是安全。你给 Agent 一个 shell，它就能 rm -rf /。TrustClaw 的设计思路很明确：
用它的安全表格自己感受：
| | TrustClaw | 本地 Agent |
|---|---|---|
| 部署时间 | 秒级 | 几小时配置 |
| 凭证管理 | Composio 加密托管 | 本地明文 |
| 代码执行 | 远程沙箱 | 你电脑上跑 |
| 工具集成 | OAuth，1000+ | 手动配 API Key |
| 审计日志 | 完整执行记录 | 没有 |
技术栈一览
Next.js 15 + React 19 → tRPC → Postgres (pgvector)
→ Redis (流式恢复)
→ Vercel AI Gateway (LLM + embedding)
→ Composio SDK (1000+ 工具)
三层上下文管理（pruning + memory flush + summarization compaction），对话可以跑无限长。
本地开发
pnpm install
cp .env.example .env # 填 DATABASE_URL, BETTER_AUTH_SECRET, COMPOSIO_API_KEY
pnpm prisma db push # 建表（需要 Postgres + pgvector）
pnpm dev # http://localhost:3000
Telegram 集成：
# 把 bot webhook 指向你的实例
https://<你的域名>/api/telegram-webhook
说人话的总结
如果你想要一个「睡觉时帮你干活」的 AI 助手，TrustClaw 是目前门槛最低的选择。它不像 OpenClaw 那样给你一台完整的电脑，而是更安全——给你一个沙箱化的、OAuth 保护的、有审计日志的员工，而不是管理员。
Vercel Hobby 计划的注意：免费计划 cron 每天只能跑一次，函数限时 300 秒。要全功能请升 Pro。
🔐 TrustClaw: Composio's 575★ Self-Hosted AI Agent — Deploy in 1 Command, Works 24/7 While You Sleep
TrustClaw is the freshly open-sourced personal AI assistant from Composio (28k★ tool integration platform). Unlike tools that require hours of config and API key management, TrustClaw is "deploy and done":
npx @composio/trustclaw deploy
One command, ~2 minutes, and you get a 24/7 AI assistant that chats on Telegram/Web, integrates 1000+ tools (Gmail, GitHub, Slack, Notion, Stripe…), has long-term memory via pgvector, and runs scheduled tasks on autopilot.
No Anthropic or OpenAI API keys required — LLM calls route through Vercel AI Gateway.
What Problem Does It Solve?
The #1 pain point with running AI agents locally is security. Give an agent a shell and it can rm -rf /. TrustClaw's design is deliberate:
| | TrustClaw | Vanilla Local Agents |
|---|---|---|
| Setup | Seconds | Hours of config |
| Credentials | Encrypted, by Composio | Plaintext in config files |
| Code Execution | Remote sandbox | Your local machine |
| Integrations | OAuth, 1000+ apps | Manual API key per app |
| Audit Trails | Full action log | None |
Tech Stack at a Glance
Next.js 15 + React 19 → tRPC → Postgres (pgvector)
→ Redis (resumable streams)
→ Vercel AI Gateway (LLM + embeddings)
→ Composio SDK (1000+ tools)
3-layer context management (pruning + memory flush + summarization compaction) lets conversations run indefinitely.
Local Dev
pnpm install
cp .env.example .env # fill DATABASE_URL, BETTER_AUTH_SECRET, COMPOSIO_API_KEY
pnpm prisma db push # apply schema (Postgres + pgvector)
pnpm dev # http://localhost:3000
Telegram integration:
# Point your bot webhook at:
https://<your-domain>/api/telegram-webhook
The Bottom Line
If you want an AI assistant that "does things while you sleep", TrustClaw is the lowest-friction option out there. Unlike OpenClaw which gives the agent a full computer, TrustClaw is more secure — it gives the agent a sandboxed employee with OAuth guardrails, not an admin shell.
Heads-up for Vercel Hobby plan: free tier cron is limited to once daily, functions capped at 300s. Upgrade to Pro for full features.