AI Providers

Kuse Cowork supports multiple AI providers, giving you flexibility in choosing the right model for your needs.

Supported Providers

Official API Providers

These providers require API keys from the respective companies.

Anthropic Claude

The default and recommended provider for most tasks.

‍

Configuration:

Base URL: https://api.anthropic.com
Auth: x-api-key header

Get your API key at console.anthropic.com

OpenAI

Support for GPT models including the latest GPT-5 series.

‍

GPT-5 Responses API

GPT-5 models use OpenAI's new Responses API format, which is automatically detected and handled.


Get your API key at platform.openai.com

Google Gemini

Google's latest AI models with thinking capabilities.

‍

Special Features:

• Thinking/reasoning mode with thoughtSignature support
• Function calling with thought signatures

Get your API key at ai.google.dev

Minimax

Advanced Chinese language model provider.

Local Inference

Run models locally for privacy and offline use.

Ollama

The easiest way to run local models.

Setup:

# Install Ollama
curl -fsSL https://ollama.ai/install.sh | sh

# Pull a model
ollama pull llama3.3:latest

Available Models:

‍

Configuration:

Base URL: http://localhost:11434
Auth: None required

LocalAI

OpenAI-compatible local inference server.

Base URL: http://localhost:8080
Auth: None required

vLLM / SGLang / TGI

High-performance inference servers:

Aggregation Services

Access multiple models through a single API.

OpenRouter

Access 100+ models through one API.

Get your API key at openrouter.ai

Groq

Ultra-fast inference with specialized hardware.

Get your API key at console.groq.com

Together AI

Cloud inference for open-source models.

DeepSeek

Chinese AI provider with strong coding models.

SiliconFlow

Cloud inference service with Chinese model focus.

Provider Configuration

Switching Providers

1. Open Settings (⚙️)
2. Select provider from the dropdown
3. Enter API key (if required)
4. Select model
5. Click "Test Connection"

API Key Storage

API keys are stored in:

~/.kuse-cowork/settings.db

Keys are:

• Stored locally only
• Never sent to third parties
• Associated with specific providers

Per-Provider Keys

You can configure different API keys for each provider:

{
  "providerKeys": {
    "anthropic": "sk-ant-...",
    "openai": "sk-...",
    "openrouter": "sk-or-..."
  }
}

When switching models, the appropriate key is automatically selected.

Custom Providers

OpenAI-Compatible Endpoints

Connect to any OpenAI-compatible API:

1. Select "Custom Service" as provider
2. Enter base URL
3. Configure authentication
4. Enter model ID

Example: LM Studio

Base URL: http://localhost:1234/v1
Auth: None
Model: local-model

Enterprise Deployments

For Azure OpenAI or self-hosted deployments:

Base URL: https://your-deployment.openai.azure.com
Auth: Bearer token
Model: your-deployment-name

Reasoning Models

Some models have special requirements:

Temperature Restrictions

The following models don't support custom temperature:

• OpenAI: o1-*, o3-*, gpt-5*
• DeepSeek: deepseek-reasoner

Temperature is automatically ignored for these models.

Extended Thinking

Some models support extended thinking/reasoning:

• Gemini 3: Uses thoughtSignature for function calling
• Claude: Uses extended thinking mode

Best Practices

Choosing a Provider

Cost Management

• Use smaller models for simple tasks
• Use local models for development/testing
• Monitor usage through provider dashboards

Performance Tips

• Groq offers fastest cloud inference
• Ollama is fastest for local (if you have GPU)
• Use streaming for better UX

Troubleshooting

Connection test fails

1. Verify API key is correct
2. Check base URL format
3. Ensure network connectivity
4. Check provider status page


Model not found

1. Verify model ID spelling
2. Check if model is available in your plan
3. For Ollama, ensure model is pulled


Rate limit errors

1. Reduce request frequency
2. Upgrade provider plan
3. Use multiple provider keys


Next Steps

• Agent System - Learn how the agent uses models
• Tools - Understand tool execution
• Configuration - Detailed settings

‍