Hugging Face OpenClaw Migration Guide

Hugging Face has introduced a guide to help users migrate their OpenClaw, Pi, or Open Code agents to open-source models following Anthropic's decision to limit Claude model access for Pro/Max subscribers in open agent platforms.

Migration Options for OpenClaw Agents

Users can restore agent functionality by choosing between hosted open models via Hugging Face Inference Providers or running models locally on their own hardware.

Hugging Face Inference Providers

Inference Providers offer the fastest route to restoring agent capabilities, particularly for users who lack the necessary hardware to run models locally. This platform routes requests to providers of open-source models.

Setup Process:

  1. Create a Hugging Face token in the settings.
  2. Initialize the connection using the command: openclaw onboard --auth-choice huggingface-api-key.
  3. Select a model when prompted. Hugging Face recommends GLM-5 due to its high performance on the Terminal Bench.

Configuration: Users can update their model by specifying the repo_id in the OpenClaw configuration file:

{
  "agents": {
    "defaults": {
      "model": {
        "primary": "huggingface/zai-org/GLM-5:fastest"
      }
    }
  }
}

HF PRO subscribers receive $2 in free monthly credits toward Inference Providers usage.

Local Model Setup with llama.cpp

Running models locally provides full privacy, zero API costs, and the removal of rate limits.

Installation:

  • macOS/Linux: brew install llama.cpp
  • Windows: winget install llama.cpp

Execution: Users can start a local server with a built-in web UI using a command such as: llama-server -hf unsloth/Qwen3.5-35B-A3B-GGUF:UD-Q4_K_XL

For this specific model (Qwen3.5-35B-A3B), 32GB of RAM is recommended. Users can browse thousands of GGUF models on the Hugging Face Hub to find one compatible with their hardware.

OpenClaw Integration: To connect OpenClaw to a local llama.cpp instance, use the following configuration command:

openclaw onboard --non-interactive \
   --auth-choice custom-api-key \
   --custom-base-url "http://127.0.0.1:8080/v1" \
   --custom-model-id "unsloth-qwen3.5-35b-a3b-gguf" \
   --custom-api-key "llama.cpp" \
   --secret-input-mode plaintext \
   --custom-compatibility openai

Verification can be performed via curl: curl http://127.0.0.1:8080/v1/models.

Comparison: Hosted vs. Local Deployment

Feature Hugging Face Inference Providers Local Setup (llama.cpp)
Speed of Setup Fastest Slower
Cost API-based (with some free credits) Zero API costs
Privacy Provider-hosted Full local control
Hardware No local hardware required Requires compatible hardware
Rate Limits Subject to provider limits No rate limits

Sources