OpenJetGet Started

cloud coding agent

Run your coding agents in the cloud, on a GPU you rent by the hour

Hosted cloud agents bill per task or per token and cap your usage. A rented GPU bills by the hour, however many tokens your agents generate.

Deploy OpenJet's template on Vast.ai or Runpod and you get an RTX 3090 with OpenJet's engine and Qwen3.8-27B already installed. SSH in and run OpenJet's agent, or point Claude Code, Codex, OpenCode or OpenClaw at the machine's endpoint. Leave agents working while your laptop is closed, and delete the machine when you're done.

Why local wins

  • No per-token billing. You pay the provider's hourly rate for the GPU, about $0.20 an hour for an RTX 3090, with no usage caps or rate limits.
  • Fast on one GPU. OpenJet's engine decodes Qwen3.8-27B at 121 tokens per second on an RTX 3090 with thinking on, about twice llama.cpp's speed, and 96 tok/s with 98K tokens in context.
  • Several agents at once. OpenJet's engine decodes up to 4 agents together in one pass over the weights, 424 tok/s in total on short coding prompts.
  • Bring your own agent. The server speaks the Anthropic Messages, OpenAI Responses and OpenAI Chat Completions APIs, so Claude Code, Codex, OpenCode and OpenClaw connect directly.

Common deployment paths

  • Leave long refactors running overnight without paying per token.
  • Give each of several agents its own task on one rented GPU.
  • Run Claude Code against an open model when you hit your plan's usage limits.
  • Try a 27B coding model without buying a 24 GB card.

Frequently asked questions

What does a cloud coding agent cost with OpenJet?
Only the provider's hourly GPU rate, about $0.20 an hour for an RTX 3090 on Vast.ai or $0.22 on Runpod Community Cloud, for as long as the machine exists, plus their storage and bandwidth charges.
Which agents can I use?
OpenJet's own terminal agent runs on the machine. Claude Code, Codex, OpenCode and OpenClaw connect to the machine's endpoint with the API token it prints in its log.
Where does my code run?
On the machine you rent. It doesn't go to a model API, but marketplace hosts can be third parties, so check the provider's terms.

Explore OpenJet

Embed a local model with the Python SDK, run Qwen3.8-27B as a local coding agent, compare local coding-agent alternatives, read the docs, or go back to the homepage.