Model Providers

Docker Agent supports multiple AI model providers. Choose the right one for your use case, or use multiple providers in the same configuration.

Supported Providers

• OpenAI — GPT-5, GPT-5-mini, GPT-4o. The most widely used AI models.
• Anthropic — Claude models for coding and analysis. See Claude access paths for protocol, host and model-specific limits.
• Google Gemini — Gemini 2.5 Flash, Gemini 3 Pro. Fast and cost-effective.
• AWS Bedrock — access Claude, Nova, Llama, and more through AWS infrastructure.
• Docker Model Runner — run models locally with Docker. No API keys, no costs.
• Local Models — run Ollama, vLLM, or LocalAI locally. No API key required.
• Nativ — run MLX models locally on Apple Silicon through a custom provider.
• Provider Definitions — define reusable provider configurations with shared defaults for any provider type.

Quick Comparison

Provider Credentials

Use this table to find a built-in provider's config key and authentication method. For ways to supply credentials without embedding secrets in YAML, see Managing Secrets. Individual provider pages cover setup and provider-specific overrides.

Additional Built-in Providers

The credential table includes built-in aliases as well as the core providers. Select one using the inline form below:

# Use built-in providers inline
agents:
  root:
    model: mistral/mistral-large-latest

Using Multiple Providers

Different agents can use different providers in the same configuration:

models:
  claude:
    provider: anthropic
    model: claude-sonnet-5
    max_tokens: 64000
  gpt:
    provider: openai
    model: gpt-5.6-sol
  local:
    provider: dmr
    model: ai/qwen3

agents:
  root:
    model: claude # coordinator uses Claude
    sub_agents: [coder, helper]
  coder:
    model: gpt # coder uses GPT-5
  helper:
    model: local # helper runs locally for free

Built-in provider IDs follow the models.dev catalogue. The legacy IDs fireworks, together, moonshot, and opencode-zen remain accepted as fireworks-ai, togetherai, moonshotai, and opencode, respectively. New configurations use the canonical IDs; older Docker Agent binaries may not recognize them. API model names, URLs, and environment variables are unchanged. Custom provider definitions take precedence over built-in IDs.