DeepSeek

DeepSeek's advanced language models with strong reasoning capabilities and tool use support, accessible via OpenRouter.

Available Clients

DeepSeekOpenRouterClient (OpenRouter API)

Environment Setup:

Set the OPENROUTER_API_KEY environment variable

Environment Variables

export OPENROUTER_API_KEY="your-key"

or add it to the .env file.

Configuration:

Agent Configuration

main_agent:
  llm: 
    provider_class: "DeepSeekOpenRouterClient"
    model_name: "deepseek/deepseek-chat-v3.1"  # Available DeepSeek models via OpenRouter
    async_client: true
    temperature: 0.3
    top_p: 0.95
    min_p: 0.0
    top_k: -1
    max_tokens: 32000
    openrouter_api_key: "${oc.env:OPENROUTER_API_KEY,???}"
    openrouter_base_url: "${oc.env:OPENROUTER_BASE_URL,https://openrouter.ai/api/v1}"
    openrouter_provider: null # You can specify the provider to use
    disable_cache_control: false
    keep_tool_result: -1
    oai_tool_thinking: false

Usage

Example Command

# Run with DeepSeek v3.1 Chat (OpenRouter) on example dataset
uv run main.py common-benchmark --config_file_name=agent_llm_deepseek_openrouter output_dir="logs/test"

The agent_llm_deepseek_openrouter.yaml configuration file provides a ready-to-use setup with the example dataset benchmark, while agent_gaia-validation_deepseek.yaml is an setup for the GAIA-Validation benchmark with main agent and sub agent configured.

---

Documentation Info

Last Updated: October 2025 · Doc Contributor: Team @ MiroMind AI