codehour

Your coding models, by the hour,
on your own machines.

codehour rents a GPU the moment you start coding and stops paying when you stop, serves it to any coding agent through one URL, and puts the whole thing in your pocket: every session, every approval, every dollar.

One URL for any agent

An OpenAI-compatible endpoint that serves your pod when it is up and a cheap hosted model otherwise. DeepSeek Harness first, Pi too — the model id never changes.

Approve from your phone

When an agent asks to run something risky, the question reaches you — Telegram buttons or the app — with a full audit trail. Nothing runs until you say so.

Pay for hours, not months

Idle pods stop themselves. A daily budget refuses to overspend. You see the cost of every session and what the alternative would have been.

A bot your clients can message

Hand a Telegram bot to a client: their messages become tasks, an agent does the work in that client's own workspace, the answer comes back in the chat.

Self-hosted · open source

Run it yourself

  • Your box, your RunPod and API keys, your data. Nothing leaves it unless you say so.
  • One Docker image or pip install; Cloudflare Tunnel + Access to reach it from anywhere.
  • MIT licensed. The source opens with the first public release.
GitHub — opening soon
codehour cloud · managed

Let us run it for you

  • We host the control plane; you add your own machines and your own RunPod account.
  • A flat monthly price per machine. GPU time stays between you and RunPod — we never resell it.
  • Updates, backups and uptime are ours to worry about.
Join the waitlist

Private beta first. Say what machines and models you use.