Hetzner Inference as a Provider for OpenCode

Use the Hetzner Inference API as a custom model provider in opencode.

Experimental — The Hetzner Inference API is currently an experiment and free of charge. Pricing, rate limits, and available models may change. This project is not affiliated with or endorsed by Hetzner.

Setup

Security note — The token is stored as plain text in ~/.config/opencode/hetzner-token. Anyone with filesystem access to your machine can read it. opencode's {file:...} substitution is a convenience, not encryption.

1. Get an API token from experiments.hetzner.com

2. Save your token globally:

echo "your_hetzner_token" > ~/.config/opencode/hetzner-token

3. Copy opencode.json into ~/.config/opencode/opencode.json (merge with any existing config).

4. (Optional) To default to a Hetzner model in a specific project, add to its opencode.json:

{ "model": "hetzner/deepseek-v4-flash" }

5. Restart opencode — Hetzner models appear in /models everywhere.

Available models

Model ID	Description	Context
hetzner/deepseek-v4-flash	DeepSeek V4 Flash (304B MoE, 13B active)	512K
hetzner/glm-5.2	GLM 5.2 (744B MoE, 40B active)	512K
hetzner/kimi-k2.7-code	Kimi K2.7 Code (1T MoE, 32B active)	262K
hetzner/qwen3.6-35b	Qwen 3.6 35B A3B (35B MoE, 3B active)	262K

How it works

The global config defines hetzner as a custom provider using @ai-sdk/openai-compatible pointed at the Hetzner Inference API. The token is read from ~/.config/opencode/hetzner-token via {file:hetzner-token} — resolved relative to the config file directory. No env vars needed, no per-project setup.