OpenAI

OpenAI provides the GPT model family. AgentXchain connects via the api_proxy adapter using the Chat Completions API.

Which adapter?

api_proxy with provider: "openai" — AgentXchain sends governed turn prompts to OpenAI's Chat Completions endpoint.

Prerequisites

• An OpenAI API key — get one from platform.openai.com
• OPENAI_API_KEY set in your environment
• agentxchain CLI installed

Configuration

{
  "runtimes": {
    "gpt-dev": {
      "type": "api_proxy",
      "provider": "openai",
      "model": "gpt-5.4",
      "auth_env": "OPENAI_API_KEY"
    }
  },
  "roles": {
    "dev": {
      "runtime": "gpt-dev",
      "mandate": "Implement features and fix bugs",
      "authority": "proposed"
    }
  }
}

Available models

Model	Best for
gpt-5.4	Most capable — complex coding, architecture
gpt-5.3-codex	Optimized for code generation
gpt-5.2-codex	Code generation, lower cost
gpt-oss-120b	Open-weight, self-hostable
gpt-oss-20b	Lightweight open-weight model

Codex-family models

For coding-specific tasks, use the Codex-optimized models:

{
  "runtimes": {
    "codex-impl": {
      "type": "api_proxy",
      "provider": "openai",
      "model": "gpt-5.3-codex",
      "auth_env": "OPENAI_API_KEY"
    }
  }
}

Custom endpoint (Azure OpenAI)

For Azure OpenAI Service deployments:

{
  "runtimes": {
    "azure-dev": {
      "type": "api_proxy",
      "provider": "openai",
      "model": "your-deployment-name",
      "auth_env": "AZURE_OPENAI_API_KEY",
      "base_url": "https://your-resource.openai.azure.com/openai/deployments/your-deployment/chat/completions?api-version=2024-02-01"
    }
  }
}

Verify the connection

export OPENAI_API_KEY="sk-..."
agentxchain connector check

Budget configuration

To supply custom cost rates for models not in the bundled defaults:

{
  "budget": {
    "max_cost_per_run": 10.00,
    "cost_rates": {
      "gpt-5.4": { "input_per_million": 5.00, "output_per_million": 15.00 }
    }
  }
}

Gotchas

• Model naming: OpenAI model names change frequently. Check the OpenAI models page for current model IDs.
• GPT-OSS models: The open-weight GPT-OSS models can also be self-hosted via Ollama or other inference platforms. See the Ollama guide or Groq guide.
• Rate limits: OpenAI applies tiered rate limits. For governed runs with many sequential turns, you may hit rate limits on lower tiers.