v0.1.0-alpha.8: first public release

First public release of dsh-hf-spaces: run open-weight LLMs on your own GPU virtual machine on Hugging Face (a Space) and use them inside DeepSeek Harness like any other model. The plugin starts the Space when you send a message, shows the start progress in the conversation header, and pauses it when idle or at a time limit, so the GPU bills only while you use it.

Alpha. Tested on DeepSeek Harness Desktop 0.2.1-rc.3 (Windows) with a private A100 Space serving a 27B GGUF model through llama.cpp llama-server: ready in about 3 minutes, an 18-minute session cost about US$ 0.75.

What is in it

• Your Space's models in the Harness model selector, with only the reasoning-effort levels your server accepts.
• Start on demand from the chat, with the conversation resuming by itself when the model is ready.
• Stop rules: idle time, a per-start time limit with a warning and a one-click extension, Stop and /space off.
• Interface: a one-line header indicator, a "Hugging Face Spaces" sidebar tab (progress, spending today and this month, recent starts, run log, credentials), notices, and a setup page with every check and value.
• Records of recent starts and their cost. No account balance: Hugging Face does not expose it through its API.

Before you install

• You set up the Space yourself: GPU hardware, an OpenAI-compatible inference engine (for example llama.cpp), the weights and the sleep time. See What you set up, and what the plugin does.
• A paid Hugging Face plan is needed for GPU Spaces; the PRO plan is recommended. See Requirements.
• Read Responsible use. You choose the models, including ones with fewer refusals; the plugin adds no content filter.

Install

Download dsh-hf-spaces-0.1.0-alpha.8.tgz below and follow Installation, step by step. The token step is always yours: never paste a token in a chat.

Full history in CHANGELOG.md.