TanStack AI runs coding agents in a sandbox now 📦 Agents: Grok build, Claude Code, Codex, OpenCode, or any ACP agent Sandboxes: Cloudflare, Docker, Local, Daytona or Vercel sandboxes Fully configurable sandboxes with skills, mcps and more! 🚀

Run your agents in sandboxes on @Cloudflare @daytonaio @vercel and @Docker

Vercel Ai vs Tanstack Ai for those interested:

Coo!

love it!

sandboxes are just wrappers for 'i have no idea what to do next' until you do

Cool

This is fantastic, I'm also looking into ways to sandbox agent executions on user owned hardware (or cloud), but this might've just solved the hard part for my use case, thanks for this!

no big deal

not fair i was just in the middle of building it....

this is fantastic - chipping away at the overpriced crap that is vercel and `ai-sdk` one feature at a time.

Capital efficiency is 𝐀𝐛𝐨𝐮t to change Cc:@MultichainZ_

Great work, please keep em coming

the isolation is the part people skip until two agents race the same resource. a flock on my browser lock caught that once, two posters raced and left half-finished work live.

nice. the thing that bit us with agent sandboxes: under bubblewrap any tool that reads its own /proc just fails silently, so the agent's compiler calls die with no error. picking the sandbox turned out to be half the work. do the cloudflare/daytona ones dodge that?

@tan_stack Huge thanks for the sandbox feature We used TanStack AI Sandboxes with OpenCode to let a Slack agent inspect a repo, write code, run tests, and open a GitHub PR. Would love for you to check it out: Demo :

The caveat is ownership. Agents can speed the work, but teams still need audit trails for what changed and why.