Cloud runner for autonomous dev agents
One control plane for every coding agent.
Run Claude Code, Codex, opencode and Gemini CLI in isolated cloud workspaces. Close the lid, approve from your phone, wake up to a pull request.
Works with
The agents you already use, and the tools they need
Agents
Connectors
Why Overnite?
The other ways to run an agent, and what they leave out
Claude Code on the web
Claude Code in Anthropic's cloud, started from the browser or the app.
One vendor's agent.
Codex Cloud
Codex in OpenAI's cloud, working on tasks in parallel.
One vendor's agent.
GitHub Copilot coding agent
Copilot picks up an issue and opens a pull request inside GitHub.
One vendor's agent, one way to start it.
A VPS and tmux
Any agent you like, on a machine that stays on.
Your machine to maintain, nothing on your phone, no isolation between agents.
Overnite
- Claude Code, Codex, opencode and Gemini CLI in the same repository, side by side.
- Agents start and supervise each other with
ovn. - Your own API key or subscription for every one of them.
- When one needs a decision, you approve it from your phone.
- Every workspace isolated in its own gVisor sandbox.
Architecture
How it works
Connect a repo
GitHub login, pick a repo, we boot an isolated machine with it cloned.
isolated sandbox / 2 vCPU
Spawn agents
Claude Code, Codex, opencode side by side, each in its own git worktree. Your API keys, our compute.
independent worktree isolation
Walk away
Agents keep working. When one needs a decision, your phone buzzes. Answer, go back to sleep.
APNs & WebPush webhooks
System Specifications
Built for agents that run for hours
Machines that never sleep
Not tmux on your laptop. Real isolated machines in the EU that keep running when you don't.
You never hunt for the stuck one
Every agent is marked working, needs input or done. The one that needs you is the only thing that blinks.
Approve from your phone
Push notification, one tap, back to bed.
Scheduled runs
"Every night at 02:00: update dependencies, run tests, open a PR." Agents on a cron.
Vendor-agnostic
Claude Code writes, Codex reviews. Mix agents in one workspace.
Agent-native API
Agents spawn and wait on each other through the same CLI you use. Compatible with the Herdr socket API.
Scheduled run
A night nobody had to watch
ovn spawn codex --task "update dependencies, run the tests, open a PR"
[02:00:41] Codex started in its own worktree (2 vCPU / 4 GB)
[02:07:12] 23 packages updated · 412 tests passed
[02:07:30] PR #143 opened: "chore: nightly dependency update"
[02:37:30] Idle for 30 minutes: workspace asleep, compute stopped
Plans
Simple pricing
Bring your own key or subscription — we never see your LLM bill.
- Join the waitlist
Free
$0 /mo
- 1 workspace
- 2 agents
- 10 compute hours/mo
Recommended
Request Pro accessPro
$19 /mo
- 3 workspaces
- 5 agents per workspace
- 100 compute hours/mo
- Request Team access
TeamComing soon
$49 /seat/mo
- 10 workspaces
- 5 agents per workspace
- 300 compute hours per seat/mo
A compute hour is one hour of a workspace being awake, regardless of how many agents run inside it.
A sleeping workspace is free, its disk included, up to 10 GB.
Every plan: isolated gVisor workspaces in the EU, push notifications and approvals from your phone, scheduled runs, and connectors.
Paid plans open after the closed beta; until then every button leads to the waitlist. Overage $0.25 per compute hour. LLM usage is billed by your provider — we never mark up tokens.
Information
Frequently asked questions
Do I need to install anything?
No. You start agents from the web cockpit, and answer them there or from your phone. Inside a workspace, agents and terminals also have ovn, a CLI for starting and supervising other agents.
Whose API keys are used?
Yours. Agents run with your own API key or your own signed-in subscription: Anthropic, OpenAI, Google, Moonshot for Kimi, and xAI for Grok Code in opencode. You pay model providers directly, with zero markup from us.
Where do my repos and keys live?
On our servers in Finland (EU). Keys are encrypted at rest with AES-256-GCM, handed to an agent only in its own environment when it starts, and never logged. Your workspace disk persists while it sleeps so you don't re-clone; delete the workspace and it's gone.
Can I use my Claude Max subscription?
Yes: sign Claude Code in inside a workspace once and it carries over. Make sure your provider's terms allow it for how you use it - that is your call, not ours. API keys avoid subscription rate-limit stalls on long autonomous runs.
How is this different from Herdr?
Herdr runs on your machine; Overnite runs on ours so you can close your laptop and walk away.
Ship it overnite.
Run agents while you sleep. Review in the morning.