Epic: API Translation

[shaneutt]

Bidirectional API format translation across inference providers.
Workloads send requests in one format (OpenAI by default)
and the gateway translates to the target provider's native
format transparently. Responses are normalized back.

Scope

• Bidirectional request/response rewriting per provider
• Streaming SSE format conversion across providers
• Model name virtualization (aliases per-backend)
• Unified client API surface
• Provider-specific field passthrough

Providers

• OpenAI (passthrough, reference format)
• Anthropic Messages API (bidirectional)
• AWS Bedrock (Converse + InvokeModel)
• Google Vertex AI
• Azure OpenAI (deployment endpoint rewriting)

Implementation tracking

OpenAI Responses ↔ Chat Completions

• Filter 16: responses_to_chat_completions #35 — Responses request and finite Chat Completions response conversion
• Filter 17: chat_completions_to_responses_sse #36 — Chat Completions streaming response → Responses SSE

Anthropic Messages ↔ OpenAI

• Anthropic Messages API #103 — classification, routing, validation, and bidirectional translation
• feat(anthropic): support MCP toolsets in the Messages API #638 — Anthropic MCP connector and mcp_toolset compatibility

Related

• Epic: AI Inference #70 — AI Inference provider abstraction framework
• Epic: Responses API and Agentic Loop Orchestration #93 — Responses API and agentic-loop orchestration