Integrations

Mixen speaks the OpenAI protocol (Chat Completions and Responses) and Anthropic (/v1/messages) — so it plugs in anywhere a custom base URL can be set. The scheme is always the same: https://api.mixen.ai/v1 and an mxn-… key from the cabinet.

Ready-made guides

Client	Type
Claude Code	Terminal agent (Anthropic protocol)
Codex CLI	OpenAI agent (Responses API)
Cursor, Cline, Roo Code, Kilo Code, Continue	IDEs and editors
OpenCode	Terminal agents
Qwen Code	Terminal agent (OpenAI protocol)
Claude Desktop	Anthropic desktop app (gateway)
OpenClaw, Hermes Agent	Standalone agents
LangChain, Vercel AI SDK	Code libraries
Open WebUI, n8n, Dify	Chat panels and no-code platforms
RAG	A search system over your own documents (guide)

Tool calling

Tool calling (tools, tool_choice) is supported on all three chat endpoints, and multi-round flows work everywhere. Each protocol signals a call differently:

Endpoint	Call signal	Tool reply
/v1/chat/completions	finish_reason: "tool_calls"	a role: "tool" message
/v1/messages (Anthropic)	stop_reason: "tool_use"	a tool_result block
/v1/responses (Codex CLI, OpenAI Agents SDK; stateless)	function_call items in output	function_call_output items

See Claude Code and Codex CLI for details.

Truncated replies

When a reply hits the length ceiling, the response says so — SDKs and agents use this to continue generating instead of treating the thought as finished:

Endpoint	Truncation signal
/v1/chat/completions	finish_reason: "length"
/v1/messages	stop_reason: "max_tokens"
/v1/responses	status: "incomplete", incomplete_details.reason: "max_output_tokens"

Not supported

• The Assistants API — OpenAI itself has deprecated it in favor of the Responses API; Mixen speaks the modern stack directly: Chat Completions, Responses and Anthropic Messages.
• Batch API and fine-tuning — also not available; model fine-tuning is not provided.

Web search is available though — not as a tool, but by picking a model with type: search, see Chat & streaming.