Claude OneClick —
Space to toggle
🌐 Provider
OneClick
⌨ Claude Code
checking…

—

How to use this tab: set the friendly label, the provider's base URL (e.g. https://api.deepseek.com), paste your API key, and pick the request format. OpenAI-compatible is right for almost every alternative provider; the local proxy will translate Claude Code's Anthropic-format calls. Pick Anthropic (direct) only when the upstream itself speaks /v1/messages.

The provider's API host. With or without /v1 at the end — both work. Examples: https://api.deepseek.com, https://integrate.api.nvidia.com, https://openrouter.ai/api/v1, http://localhost:11434.

Saving propagates to every preset in the same provider group (one API key = one provider account). If a sibling preset isn't picking it up, click Save key again with the same key — that re-runs propagation across siblings.

Pick OpenAI-compatible if the provider's docs show /v1/chat/completions. Pick Anthropic only when the upstream natively serves /v1/messages in Anthropic's request shape — most third-party providers don't.

Models

Click Refresh to pull the live list from the provider (uses GET /v1/models, or Ollama's /api/tags). For NVIDIA NIMs that's everything they host right now — Llama, Nemotron, DeepSeek V4 Pro/Flash, GLM 4 / 4.5, Phi, Mistral, etc. The dropdown is also a free-text input.

No model list loaded yet — click Refresh.

Aliases

Map names users type (e.g. claude-sonnet-4) to upstream model ids.

Defaults applied only when Claude Code's request omits them. Leave blank to pass the request through unchanged. top_k is silently dropped for OpenAI-style upstreams (they don't accept it).
Extra HTTP headers added to every request to the upstream provider. Common ones:
  • HTTP-Referer + X-Title — required by OpenRouter for proper attribution
  • accept: application/json — required by some NVIDIA NIMs endpoints
  • custom X-… headers your enterprise gateway requires
Power-user knobs. System prompt prefix/suffix is bolted onto the system message before forwarding (handy for jailbreak prompts or persona text). Disable streaming falls back to a single non-streamed response — try this first if you see corrupted output. Pass cache_control only matters on providers that support Anthropic's prompt-cache markers.
Reasoning

Works with any reasoning-capable upstream — OpenAI o-series and DeepSeek V4 Pro/Flash get the standard reasoning_effort field; for upstreams that use a different name, use Extra request body below with the literal {{effort}} placeholder.

lowmediumhigh
medium

JSON object merged into every outbound request. The literal string {{effort}} in any value is replaced by the slider's current level when reasoning is ON.

One-config-fits-all. Copy the active preset into any other AI coding tool — same provider, same key, same model. Each snippet is generated from your current Setup tab values.

Roo Code (VSCode)

Add to VSCode settings.json.


          
        

        

          
Cline (VSCode)

          
Add to VSCode settings.json.

          

          
        

        

          
Continue.dev

          
Drop into ~/.continue/config.json under models.

          

          
        

        

          
Cursor

          
Cursor → Settings → Models → "Add custom OpenAI-compatible API".

          

          
        

        

          
Codex / OpenAI CLI

          
Shell env vars (works for the OpenAI Python SDK and most other OpenAI-format tools).

          

          
        

        

          
Aider

          
Shell env vars + a --model flag.

          

          
        

        

          
Generic shell env

          
For any tool that reads OPENAI_* or ANTHROPIC_* env.

          

          
        

      

    


    

    

      

        Live tail of ~/.config/claude-oneclick/proxy.log. If
        Claude Code is misbehaving, click Refresh to see the most
        recent upstream calls and any error responses.
      

      

        
        
        
      

      

    

  





  

    
Are you sure?

    

    

      
      
    

  




  

    
Settings

    
Global configuration. Proxy port change applies on save (the proxy restarts). UI port change takes effect the next time you run claude-oneclick ui.

    

    

    

      
    

    

      
    

    

    

    

    

    

    
API key storage

    

    
—


    

    
Surfaces

    
Per-surface routing. The shell rc + system env always apply globally; these are extras.

    

    
—

    
For per-surface env without changing global state, use the CLI: claude-oneclick run cli, claude-oneclick run code, claude-oneclick run desktop.


    

    
Updates

    

      
      —
      
    

    

      
      Click "Check for updates" to query GitHub.
      
    

    
Updates compare your install's commit SHA to github.com/xhashmi/claude-oneclick. Works for both git clone installs (uses git pull) and ZIP installs (downloads the latest branch ZIP and extracts in place).