Roadmap

Anthropic /v1/messages shape

Meridian currently speaks the OpenAI chat-completions shape exclusively. Native support for the Anthropic Messages API is on the roadmap — no ETA yet, but the design is locked.

Current state

  • All providers (OpenAI, Anthropic, Gemini, Groq) are normalized through the OpenAI chat-completions wire format.
  • Anthropic models work today — Meridian translates system/user/assistant messages and tool-use blocks internally.
  • Streaming, stop reasons, and token counts are mapped to the OpenAI shape in responses.

Planned native shape

  • First-class POST /v1/messages endpoint alongside the existing chat-completions path.
  • Full support for the Anthropic content-block array shape: text, image, tool_use, tool_result.
  • Native streaming via server-sent events matching the Anthropic event types: message_start, content_block_delta, message_delta, message_stop.
  • Prompt caching headers (ephemeral) and extended thinking budget controls.

When the native shape ships, existing OpenAI-shaped traffic will continue to work without changes. The Messages endpoint is additive — pick the shape that fits your stack. If you need this sooner, reach out via Discord.