Skip to main content
Poixe AI is compatible with 4 mainstream text chat protocols. All of them can be called through a unified entry address, but request structure and authentication strictly follow each protocol’s official definition.

​
No protocol conversion

The system does not “translate” one protocol into another. When a request calls a provider’s model using Chat Completions, the system goes directly through that provider’s officially provided Chat Completions-compatible implementation, rather than first calling with another protocol and then converting the result to Chat Completions.

​
Protocols at a glance

OpenAI Chat Completions Protocol

Compatible with the legacy ecosystem: suited to integrations built on the existing OpenAI ecosystem.

OpenAI Responses Protocol

A more unified content structure with extended support for multimodality and tool calling.

Anthropic Messages Protocol

Integrates with the Claude ecosystem: organizes input and output using the Messages structure.

Google Gemini Content Protocol

Integrates with the Gemini ecosystem: based on the generateContent / streamGenerateContent structure.

​
Compatibility strategy and boundaries

​
1. OpenAI Chat Completions (best compatibility)

  • The most classic protocol with the broadest ecosystem; almost every major model provider offers this protocol or a compatible implementation
  • For “regular chat / text generation”, it is usually sufficient
  • Poixe AI’s integration experience and compatibility handling are also the most mature on this protocol (recommended as the default choice)

​
2. OpenAI Responses (OpenAI’s newer protocol)

  • A more unified interface shape, suited to organizing more “modern” conversations and multi-turn flows
  • Expresses multimodality and tool extensions more naturally (subject to the official definition)
  • Note: it is only available when the model provider you want to call supports Responses itself (or offers a compatible implementation)

​
3. Anthropic Messages (the Claude protocol)

  • Anthropic’s official Messages structure (subject to the official definition)
  • Commonly used in scenarios that need Claude’s native capabilities (for example, its native tool/workflow capabilities)
  • Likewise limited by whether the model provider offers the Messages protocol

​
4. Google Gemini Content (the Gemini protocol)

  • Google’s official generateContent / streamGenerateContent structure (subject to the official definition)
  • Commonly used in scenarios that need Gemini’s native capabilities (for example, its native tool/workflow capabilities)
  • Likewise limited by whether the model provider offers the Content protocol
Summary: Apart from Chat Completions, the other protocols are essentially supported “as-is per the official protocol”, and what is available depends on provider and model support.

​
How to choose

  • Regular chat / text generation only, with maximum compatibility and the fewest pitfalls: prefer Chat Completions
  • You explicitly need OpenAI’s Responses ecosystem capabilities (for example, more unified multi-turn flows / extensions): use Responses
  • You explicitly need Claude’s native protocol capabilities: use Messages
  • You explicitly need Gemini’s native protocol capabilities: use Content