LM Studio Local Provider Settings

Provider Configuration

Copy
# Required: LM Studio local server settings
export ANTHROPIC_BASE_URL=http://localhost:1234/v1
export ANTHROPIC_API_KEY=not-needed

Settings File

Copy
{
  "env": {
    "ANTHROPIC_BASE_URL": "http://localhost:1234/v1",
    "ANTHROPIC_API_KEY": "not-needed"
  },
  "preferences": {
    "model": "qwen3-coder"
  }
}

LM Studio Setup

1. Install LM Studio

Download from https://lmstudio.ai for macOS, Windows, or Linux.

2. Download a Coding Model

Recommended models for coding tasks:

Model	VRAM Required	Quality
Qwen3-Coder 32B (Q4)	~20 GB	Excellent
DeepSeek Coder V2 16B (Q4)	~10 GB	Very Good
CodeLlama 34B (Q4)	~20 GB	Good
Qwen2.5 Coder 7B (Q4)	~5 GB	Decent

3. Start the Local Server

1. Open LM Studio
2. Load your chosen model
3. Go to "Local Server" tab
4. Click "Start Server" (default port: 1234)
5. Verify: curl http://localhost:1234/v1/models

Hardware Requirements

Setup	Minimum	Recommended
7B model	8 GB RAM, 6 GB VRAM	16 GB RAM, 8 GB VRAM
14B model	16 GB RAM, 10 GB VRAM	32 GB RAM, 12 GB VRAM
32B model	32 GB RAM, 20 GB VRAM	64 GB RAM, 24 GB VRAM

Apple Silicon Macs can use unified memory for both RAM and VRAM.

Limitations

• Slower than cloud APIs (depends on your GPU)
• Smaller context windows (4K-32K vs Claude's 200K)
• Lower quality for complex multi-step reasoning
• Tool use may not work reliably with all models
• Extended thinking and advanced features not supported

When to Use Local Models

• Air-gapped or classified environments
• Sensitive codebases that cannot use cloud APIs
• Offline development (travel, no internet)
• Learning and experimentation
• Reducing API costs for simple tasks