OpenAI (GPT-4)

Agent Smith supports OpenAI models with the same agentic loop and tool calling as Claude. GPT-4.1 is the recommended model.

Setup

1. Get an API key from platform.openai.com

2. Set the environment variable:

export OPENAI_API_KEY=sk-...

3. Configure agentsmith.yml:

projects:
  my-api:
    agent:
      type: OpenAI
      model: gpt-4.1

secrets:
  openai_api_key: ${OPENAI_API_KEY}

Model Routing

Use GPT-4.1-mini for lightweight tasks:

agent:
  type: OpenAI
  model: gpt-4.1
  models:
    scout:
      model: gpt-4.1-mini
      max_tokens: 4096
    primary:
      model: gpt-4.1
      max_tokens: 8192
    planning:
      model: gpt-4.1
      max_tokens: 4096
    summarization:
      model: gpt-4.1-mini
      max_tokens: 2048

Retry Configuration

OpenAI returns 429 on rate limits. The retry config works identically to Claude:

agent:
  type: OpenAI
  retry:
    max_retries: 5
    initial_delay_ms: 2000
    backoff_multiplier: 2.0
    max_delay_ms: 60000

Pricing Configuration

agent:
  pricing:
    models:
      gpt-4.1:
        input_per_million: 2.0
        output_per_million: 8.0
      gpt-4.1-mini:
        input_per_million: 0.40
        output_per_million: 1.60

Custom Endpoint

To use an OpenAI-compatible proxy or Azure OpenAI, set endpoint:

agent:
  type: OpenAI
  model: gpt-4.1
  endpoint: https://my-proxy.example.com/v1
  api_key_secret: MY_PROXY_KEY    # Resolves from secrets section

Tip

For third-party OpenAI-compatible endpoints (Groq, Together AI, vLLM, etc.), see the OpenAI-Compatible page.

Feature Comparison with Claude

Feature	OpenAI	Claude
Tool calling	Yes	Yes
Agentic loop	Yes	Yes
Multi-model routing	Yes	Yes
Prompt caching	No	Yes
Context compaction	Yes	Yes
Cost tracking	Yes	Yes

The main trade-off: OpenAI does not support prompt caching, so repeated system prompts are billed at full input price on every call. For long agentic loops, Claude with caching enabled is significantly cheaper.