@lyfuci/openclaw-hf

Complete Hugging Face Inference Providers plugin for OpenClaw — covers all HF capability areas under a single HF API token.

Capability	Provider route	Default model
LLM chat	router.huggingface.co/v1 (OpenAI-compat)	any model in your config
Image generation	hf-inference/models/<id>	black-forest-labs/FLUX.1-schnell
Memory embeddings	scaleway/v1/embeddings	Qwen/Qwen3-Embedding-8B
Audio transcription	hf-inference/models/<id>	openai/whisper-large-v3
Video generation	replicate/v1/models/<id>	Wan-AI/Wan2.2-T2V-A14B

HF Pro $2/month free tier covers all routes. No separate accounts needed.

Install

openclaw plugins install @lyfuci/openclaw-hf

Configure

Add an hf provider section to your openclaw.json:

{
  "models": {
    "providers": {
      "hf": {
        "baseUrl": "https://router.huggingface.co/v1",
        "api": "openai-completions",
        "apiKey": { "source": "env", "id": "HUGGINGFACE_HUB_TOKEN" },
        "models": [
          {
            "id": "deepseek-ai/DeepSeek-V3-0324",
            "name": "DeepSeek V3 (HF)",
            "api": "openai-completions",
            "contextWindow": 131072,
            "maxTokens": 8192,
            "compat": { "supportsUsageInStreaming": true }
          },
          {
            "id": "Qwen/Qwen3-235B-A22B",
            "name": "Qwen3 235B (HF)",
            "api": "openai-completions",
            "contextWindow": 131072,
            "maxTokens": 8192,
            "compat": { "supportsUsageInStreaming": true }
          }
        ]
      }
    }
  }
}

Note: always add "compat": {"supportsUsageInStreaming": true} to each model — without it OpenClaw forces token counts to 0 for non-OpenAI-native endpoints.

Set your token:

export HUGGINGFACE_HUB_TOKEN=hf_...

Or add it via:

openclaw auth login --provider hf --method api-key

Image generation

Switch your image model to use this provider:

{
  "agents": {
    "defaults": {
      "imageModel": { "primary": "hf/black-forest-labs/FLUX.1-schnell" }
    }
  }
}

Memory embeddings

Switch semantic memory search to this provider:

{
  "agents": {
    "defaults": {
      "memorySearch": {
        "provider": "hf",
        "model": "qwen3-embedding-8b"
      }
    }
  }
}

Video generation

Supported Wan models (HF Pro, via replicate): Wan-AI/Wan2.1-T2V-14B, Wan-AI/Wan2.1-T2V-1.3B, Wan-AI/Wan2.2-T2V-A14B, Wan-AI/Wan2.2-TI2V-5B.

Notes

• fal-ai routes (HunyuanVideo, Mochi, CogVideoX) require separate fal.ai pre-paid credits — not covered by HF Pro
• STT uses the raw bytes endpoint on hf-inference; not the OpenAI /audio/transcriptions path
• Embedding dimensions: 4096 (Qwen3-Embedding-8B via Scaleway)

License

MIT