TramAI

Provider Reference

This page summarizes the current provider modules.

tramai-anthropic

Class:

  • AnthropicProvider

Purpose:

  • Anthropic Messages API integration

Auth:

  • API key

Config shape:

tramai:
  providers:
    anthropic:
      api-key: ${ANTHROPIC_API_KEY}
      api-key-secret-ref: null
      base-url: https://api.anthropic.com

tramai-openai

Classes:

  • OpenAiProvider
  • OpenAiCompatibleProvider
  • CodexAuthFileTokenSource (experimental)

Purpose:

  • public OpenAI API integration
  • OpenAI-compatible endpoint integration

Auth modes:

  • API key
  • explicit bearer token
  • local Codex auth file, experimental
  • Spring secret references through SecretValueResolver

Config shape:

tramai:
  providers:
    openai:
      api-key: ${OPENAI_API_KEY}
      api-key-secret-ref: null
      bearer-token: null
      bearer-token-secret-ref: null

tramai-ollama

Class:

  • OllamaProvider

Purpose:

  • local Ollama chat API integration

Config shape:

tramai:
  providers:
    ollama:
      base-url: http://localhost:11434

Capability Summary

Current shared baseline:

  • request/response normalization
  • raw string operations
  • streaming support
  • tool-calling support through engine-owned orchestration
  • structured output via Tramai's schema-in-prompt pipeline
  • explicit provider registry integration
  • operation-level timeout propagation
  • engine-owned retries for retryable provider failures

Not implemented across the provider layer yet:

  • native provider-specific structured output modes
  • externalized provider-level retry-policy configuration
  • externalized provider-level timeout-policy configuration
  • bundled cloud-specific secret-store resolvers

Choosing A Provider

Use:

  • Anthropic when your app is centered on Claude models
  • OpenAI when your app is centered on OpenAI models or compatible gateways
  • Ollama for local development and self-hosted local runs
  • Gemini when your app targets Google's Gemini models
  • DeepSeek for cost-effective reasoning models
  • Bedrock when operating within AWS infrastructure
  • Azure OpenAI when your organization uses Azure

Use multiple providers when different tasks have different needs. The provider registry supports routing different models to different providers and defining fallback chains.

Previous
Next