A LiteLLM proxy solution to use Claude Code with models from the Weights and Biases inference service. You need to have LiteLLM installed or use the docker container. Easiest is to install it with `uv tool install "litellm[proxy]"` Don't worry about the fallback warnings. Either LiteLLM, W&B or the combo of both are not handling streaming respon…

cc-proxy.sh

#!/bin/bash
export WANDB_API_KEY=<your key>
export WANDB_PROJECT=<org/project>

litellm --port 4000 --debug --config cc-proxy.yaml

cc-proxy.yaml

litellm_settings:
  drop_params: true
  cache: True
  cache_params:
    type: local
  enable_preview_features: True

model_list:

  - model_name: anthropic/claude-sonnet-*
    litellm_params:
      model: openai/Qwen/Qwen3-Coder-480B-A35B-Instruct 
      api_key: "os.environ/WANDB_API_KEY"
      api_base: https://api.inference.wandb.ai/v1
      headers: 
        OpenAI-Project: "os.environ/WANDB_PROJECT"
      max_tokens: 65536
      repetition_penalty: 1.05
      temperature: 0.7
      top_k: 20
      top_p: 0.8
    model_info:
      input_cost_per_token: 0.000001
      output_cost_per_token: 0.0000015

  - model_name: anthropic/claude-opus-*
    litellm_params:
      model: openai/Qwen/Qwen3-235B-A22B-Thinking-2507
      api_key: "os.environ/WANDB_API_KEY"
      api_base: https://api.inference.wandb.ai/v1
      headers: 
        OpenAI-Project: "os.environ/WANDB_PROJECT"
      max_tokens: 65536
      repetition_penalty: 1.05
      temperature: 0.6
      top_k: 40
      top_p: 0.95
    model_info:
      input_cost_per_token: 0.0000001
      output_cost_per_token: 0.0000001

  - model_name: anthropic/claude-3-5-haiku-*
    litellm_params:
      model: openai/Qwen/Qwen3-235B-A22B-Instruct-2507
      api_key: "os.environ/WANDB_API_KEY"
      api_base: https://api.inference.wandb.ai/v1
      max_tokens: 65536
      repetition_penalty: 1.05
      temperature: 0.7
      top_k: 20
      top_p: 0.8
      headers: 
        OpenAI-Project: "os.environ/WANDB_PROJECT"
      input_cost_per_token: 0.0000001
      output_cost_per_token: 0.0000001

code-wandb.sh

#!/bin/bash
export ANTHROPIC_AUTH_TOKEN=sk-1234
export ANTHROPIC_BASE_URL=http://localhost:4000
export CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC=1

# Startin VS Code, but could also run claude here
code &