Rush Cloud - Your coding agents, isolated in the cloud.
Sandbox. Capture. Ship.
Point any of these at your repo. Each run gets its own sandbox, the full trajectory is recorded, and you get a PR with proof.
rush cloud run claude-code owner/repo --task "fix the flaky test"A real run
Claude Code reviewed a pull request in 7 minutes 32 seconds.
Pull request #2500 on the public agents repository. 187 events, 43 turns, 42 tool calls, $0.23. Captured from the console on 2026-09-10.
Why Rush Cloud
You are accountable for what an agent ships.
Rush runs it in its own sandbox, records what happened, and hands you both the pull request and the proof.
Isolated
One pod per run. Your client's code never touches another client's.
Recorded
Every tool call, every decision, every file. Hand your client the proof.
Managed
Sign in, point at a repo, run. Scaling and isolation are on us.
Open
Claude Code, Codex, or OpenCode. Bring the harness you already use.
How it works
From task to pull request, recorded end to end.
1
Task
Describe the work
2
Sandbox
Pod spins up
3
Agent
Harness runs
4
Trajectory
Every step kept
5
PR + Proof
Branch, diff, evidence
Pricing
You pay for compute and tokens. Nothing else.
Bring the Claude or Codex plan you already pay for, bring a provider key, or meter it through Rush. No seats, no per-agent subscriptions.
Bring your subscription
Claude Code · Codex
$0to Rush
Already pay for Claude Code or Codex? Connect it and agents run on the plan you already have.
Bring your API key
Anthropic · OpenAI
$0to Rush
Paste an API key from your provider. You pay them at their rates; Rush adds nothing on top.
Rush usage
No subscription or key needed
Meteredmonthly
Run agents through Rush and pay for what you use, billed monthly. Set a spend cap; runs stop at the cap.
No per-agent subscriptions, no credits, no wallet. Switch lanes any time.
From the Blog
Thoughts on the intelligence age
Why We Built Our Agent Harness in Go
Every agent framework is Python. We think that's a category error. The honest engineering case for a Go agent runtime, with the real code and the tradeoffs we won't pretend away.
Ephemeral Apps: Software That Exists for Five Minutes
Why the future of software isn't permanent installations but interfaces that spawn when you need them and dissolve when you don't.
The Siri Trap: Why Single-Agent Assistants Always Fail
Siri launched in 2011. $100B+ invested across Siri, Alexa, and Google Assistant. After 15 years, the most common use case is still 'set a timer.' The failure isn't talent. It's architecture.