Developers

One API.
Every provider.

A single endpoint for OpenAI, Anthropic, Google, Grok and OpenRouter — switch models without changing code, or let auto-routing choose. Plus a published MCP server to plug into your editor.

One call to integrate

One format, any language that speaks HTTP. model: "auto" lets the system pick the best model for the task.

index.ts
const response = await fetch(
  "https://llm.zihin.ai/api/v3/llm/public/call",
  {
    method: "POST",
    headers: {
      "X-Api-Key": process.env.ZIHIN_API_KEY,
      "Content-Type": "application/json",
    },
    body: JSON.stringify({
      query: "Resuma o último pedido do cliente 4512.",
      model: "auto", // roteamento automático entre provedores
      options: { temperature: 0.7, maxTokens: 4096 },
    }),
  }
);

const data = await response.json();
console.log(data.response);

Full request and response, structured output and multimodal: calls reference

What the API delivers

Six capabilities, all documented — no asterisks.

Multi-provider

OpenAI, Anthropic, Google, Grok and OpenRouter through the same endpoint. Switching providers is switching a string.

Auto-routing

model: "auto" and the system picks the best model for the task — including a vision-capable one if the call has an image.

Structured output

Force JSON responses with schema validation — for integrations that can't tolerate loose text.

Multimodal

Send images alongside text — automatically converted to each provider's native format.

Streaming

Real-time token-by-token responses via SSE.

Global cache

Automatic response caching — up to 40% cost reduction, zero configuration.

MCP

Your editor talks to Zihin

The official MCP server connects MCP clients to the platform: manage agents, secrets and telemetry without leaving your editor.

MCP Server documentation

published on npm

$ npx @zihin/mcp-server

Claude Desktop · Cursor · Claude Code · Codex · Windsurf

Your API Key in 30 seconds.

15-day trial, no card, 100K test tokens.