Run OpenCode in the cloud.
Point the open-source OpenCode harness at Kimi K3, DeepSeek V4 Pro, or whichever model ships next — running entirely on Oyren's remote infrastructure. No local GPU, no host to babysit.
Two commands, any model:
opencode auth login # pick Moonshot AI, DeepSeek, or any provider — paste your API key opencode # → /models → pick kimi-k3, deepseek-v4-pro, or whatever ships next
Keep the harness you trust and swap the engine as the leaderboard moves — Oyren handles the container underneath it.
From repo to draft PR
1. Connect a repo
Grant Oyren short-lived, repo-scoped access to any GitHub repository — nothing else in your account is touched.
2. Pick OpenCode + a model
Choose OpenCode as the harness and the model behind it — Kimi K3, DeepSeek V4 Pro, or anything else OpenCode supports.
3. Watch it ship a draft PR
Oyren boots an isolated cloud container, clones the repo, and the agent edits, commits, and opens a draft PR as it works.
Pick the model, not the machine
OpenCode ships no model of its own — plug in whichever one fits the task.
Kimi K3
Moonshot AI's open-weight flagship — a 2.8T-parameter mixture-of-experts model with a 1M-token context window.
DeepSeek V4 Pro
DeepSeek's frontier-tier reasoning model, built for the long, multi-step runs agentic coding needs.
+ every model OpenCode supports, with more added as they ship.
Run your first agent
Connect a repo, pick OpenCode and a model, and Oyren launches the container.
Start free