DeepSeek

DeepSeek builds high-performance coding models. AgentXchain connects via api_proxy using DeepSeek's OpenAI-compatible API.

Which adapter?

api_proxy with provider: "openai" and a custom base_url — DeepSeek's API is OpenAI-compatible.

Prerequisites

• A DeepSeek API key — get one from platform.deepseek.com
• DEEPSEEK_API_KEY set in your environment
• agentxchain CLI installed

Configuration

{
  "runtimes": {
    "deepseek-dev": {
      "type": "api_proxy",
      "provider": "openai",
      "model": "deepseek-coder-v3",
      "auth_env": "DEEPSEEK_API_KEY",
      "base_url": "https://api.deepseek.com/v1/chat/completions"
    }
  },
  "roles": {
    "dev": {
      "runtime": "deepseek-dev",
      "mandate": "Implement features and fix bugs",
      "authority": "proposed"
    }
  }
}

Minimal working example

agentxchain init --governed --template api-service --goal "Build a feedback intake API" --dir my-project -y
cd my-project
# Replace the scaffolded runtime wiring in agentxchain.json with the DeepSeek config above.
agentxchain doctor
agentxchain connector check
agentxchain connector validate deepseek-dev
agentxchain run

If you prefer the guided interactive scaffold, run agentxchain init --governed without -y, then update agentxchain.json with the DeepSeek config above before agentxchain connector check and agentxchain connector validate deepseek-dev.

Available models

Model	Best for
deepseek-v3.2	General-purpose, strong reasoning
deepseek-r2	Deep reasoning, complex architecture
deepseek-coder-v3	Optimized for code generation

Local via Ollama

DeepSeek models are open-weight and can run locally:

ollama pull deepseek-coder-v3:33b

{
  "runtimes": {
    "deepseek-local": {
      "type": "api_proxy",
      "provider": "ollama",
      "model": "deepseek-coder-v3:33b",
      "auth_env": "OLLAMA_API_KEY"
    }
  }
}

Verify the connection

export DEEPSEEK_API_KEY="sk-..."
agentxchain connector check
agentxchain connector validate deepseek-dev

Budget configuration

DeepSeek models are not in the bundled cost defaults. Add an operator override keyed by the exact runtime model string:

{
  "budget": {
    "cost_rates": {
      "deepseek-coder-v3": { "input_per_1m": 0.14, "output_per_1m": 0.28 }
    }
  }
}

Gotchas

• OpenAI-compatible: DeepSeek uses OpenAI's API format. Set provider: "openai" with a base_url override.
• Pricing: DeepSeek models are significantly cheaper than comparable proprietary models. Good for high-volume governed runs.
• Rate limits: DeepSeek applies rate limits. Check your account tier.