pi-hetzner-inference-provider

A Pi extension that registers the Hetzner Experiments Inference API as a first-class provider: OpenAI-compatible open-weight models served on Hetzner's own infrastructure. The model catalog mirrors the hetzner provider in models.dev — the same catalog OpenCode ships — so models, context/output limits, modalities and thinking levels behave exactly like they do in OpenCode.

Install

# via npm
pi install npm:pi-hetzner-inference-provider

# via git
pi install git:github.com/ipmanlk/pi-hetzner-inference-provider

# local development
pi install ./pi-hetzner-inference-provider

Setup

1. Create an API key:

   • Log in at https://experiments.hetzner.com with your Hetzner account. A payment method (credit card) must be on file before the account lets you generate a token — even though the API is free while experimental.
   • On the left-hand sidebar, under APPS, select Inference.
   • Click the Create API Token button in the top-right corner.

2. Set the token as an environment variable:

   echo 'export HETZNER_API_KEY="<your-token>"' >> ~/.bashrc
   source ~/.bashrc
   

   HETZNER_VLLM_API_KEY and HETZNER_INFERENCE_API_KEY are accepted as aliases. You can also run /login hetzner inside pi and paste the token there.

3. Start pi and pick a model:

   pi
   # /model hetzner/Qwen3.8-27B
   

Available models

Model id	Context	Max output	Modalities	Thinking
Qwen/Qwen3.6-35B-A3B-FP8	262 144	262 144	text, image	always on
Qwen3.8-27B	262 144	262 144	text, image	always on

• Qwen/Qwen3.6-35B-A3B-FP8 — open multimodal Qwen MoE (35B total / 3B active params) for agents that need vision and code. Released 2026-04-17.
• Qwen3.8-27B — dense 27B vision-language model for coding, agent tasks, and image/video understanding. Released 2026-08-14.

Both models also support tool calling and structured output.

Context and max-output figures come from models.dev limit for the hetzner provider in models.dev; modalities from its modalities. The catalog is a static snapshot — there is no live refresh, so new models appear when this package is updated.

Hetzner's Inference Experiment continues in reduced scope: it now serves small Qwen models only. The large DeepSeek, GLM and Kimi deployments were retired in August 2026 (Hetzner blog). The definitive list of served models is always the /v1/models endpoint of the API.

Thinking levels

Thinking control matches models.dev's reasoning_options for the provider:

• Qwen/Qwen3.6-35B-A3B-FP8 and Qwen3.8-27B — no reasoning_options in models.dev, so no thinking parameter is sent at all: the models reason natively and their thinking is displayed, exactly like in OpenCode.

How it works

• Provider id hetzner, base URL https://inference.hetzner.com/api/v1.
• API openai-completions (OpenAI Chat Completions) with vLLM compatibility flags: max_tokens (not max_completion_tokens), no store flag, system role only.
• Thinking is sent as OpenAI-style reasoning_effort for models with reasoning_options in the models.dev catalog. Currently none of the served models have it: no thinking parameter is sent at all, mirroring OpenCode.
• Catalog is a static snapshot of the models.dev hetzner provider (no live refresh; update the package for new models).
• Cost is tracked at $0 — the Inference API is free while in experimental status.
• A User-Agent header (pi-hetzner-inference-provider/<version>) is sent on model requests.

Troubleshooting

• Stuck on “Working…”: the Inference API is experimental and can be very slow — first requests to a model can take a minute or more while it warms up, and requests can hang when the service is under load. Retry, or switch model. If the request eventually fails, run /login hetzner to re-check the token.
• Retired models (DeepSeek, GLM, Kimi): these deployments were shut down in August 2026 and will fail with a model-not-found error. Update to the latest version of this package and pick one of the Qwen models.

License

MIT