Models

OpenAgent integrates with 30+ language model providers through a unified interface. You can mix and match providers across different agents without changing any agent logic.

Supported Providers

Cloud Providers

Chinese Providers

Local / Self-Hosted

Configuring a Provider

Via the Dashboard

1. Go to Settings → Model Providers
2. Click Add Provider
3. Select your provider from the list
4. Enter the required credentials (API key, base URL, etc.)
5. Click Test Connection, then Save

Configuration Examples

{
  "provider": "openai",
  "api_key": "sk-...",
  "models": ["gpt-4o", "gpt-4o-mini", "gpt-3.5-turbo"]
}

{
  "provider": "anthropic",
  "api_key": "sk-ant-...",
  "models": ["claude-3-5-sonnet-20241022", "claude-3-haiku-20240307"]
}

{
  "provider": "ollama",
  "base_url": "http://localhost:11434",
  "models": ["llama3.2", "mistral", "phi4", "gemma2"]
}

{
  "provider": "azure_openai",
  "api_key": "...",
  "endpoint": "https://your-resource.openai.azure.com",
  "api_version": "2024-02-01",
  "deployment_name": "gpt-4o"
}

Embedding Models

OpenAgent uses embedding models separately from chat models. Embeddings power the knowledge base search.

Model Selection Tips

For general-purpose assistants:

• gpt-4o or claude-3-5-sonnet — best balance of capability and speed
• gpt-4o-mini or claude-3-haiku — faster and cheaper for high-volume use

For document Q&A / RAG:

• Any frontier model works well; larger context windows help
• gemini-1.5-pro (1M token context) for very large document sets

For code generation and reasoning:

• deepseek-r1 or o3-mini for complex reasoning tasks
• claude-3-5-sonnet for code generation

For cost-sensitive or offline deployments:

• Ollama with llama3.2 or mistral for fully local, free inference
• gpt-4o-mini for the cheapest capable cloud model

Context Length Management

OpenAgent automatically manages context windows. When conversation history would exceed the model's context limit, it applies a sliding window strategy: the oldest turns are dropped while keeping the system prompt and recent history.

You can configure the context window size per agent to control memory vs. cost tradeoffs.