TrustClaw by Composio - Self-hosted AI agent that connects 1000+ apps on Vercel
by•
TrustClaw is a personal AI agent that you can self-host on Vercel with one command.
Powered by Composio, it connects to 1000+ apps over OAuth, runs scheduled jobs on its own, and talks to you on the web or Telegram.
It is fully open-sourced, MIT-licensed and ready for you to take over.
Replies
Product Hunt
Setup was super quick & easy using Vercel Single Click deploy but when I started using it I couldn't get it working and pages were loading blank (tools and no responses). What did I do wrong!? :(
I've spent entire weekends trying to set up self-hosted agent stuff and it always falls apart somewhere in the auth flow . the OAuth-through-Composio approach makes sense too, like the agent never actually sees my raw credentials which is the part that always made me nervous. one thing I'm genuinely unsure about though , if I'm on Vercel hobby plan, how long can a single agent task actually run before it just cuts off?
🔥 love using this!! telgram -> docs/planning feels like magic
Self-hosting the auth layer is the right call for B2B clients who won't hand over OAuth tokens to a cloud service — we ran into this constantly at our IT managed services firm. The moment you mention cloud-hosted credentials, procurement slows to a crawl. Curious how you're handling credential rotation and revocation at scale. Also any plans for a local credential store beyond env vars for teams running air-gapped environments?
Deploying a 24/7 autonomous agent with just a single npx command on Vercel is honestly such a slick developer experience. I'm really curious about the engineering behind the scenes though—since Vercel Hobby plans cap serverless functions at 5 minutes, how are you keeping complex, multi-step agent workflows from getting cut off mid-execution? Are you chunking the state shifts into separate cron intervals, or does it require a Pro plan to realistically run heavy background tasks?
Tried building an MCP-driven DevOps automation before and the security issues showed up almost immediately. Ended up sandboxing everything, containerized environments, remote git proxies, deterministic scripts for execution, basically making sure the agent never touched anything sensitive directly.
TrustClaw's architecture of OAuth-scoped access through Composio, where the agent never sees raw credentials or touches your local machine, is essentially what I was trying to piece together manually. The one-command deployment would have saved me a week!!
One thing I'm curious about though, once the agent starts accumulating long-term memory and scheduled jobs over time, how do you prevent stale context from slowly degrading decision quality?