Mira Docs

Build with Mira

Mira is a family of large language models exposed through an OpenAI-compatible HTTP API. If your code already talks to OpenAI's Chat Completions endpoint, point its base URL at https://api.vmira.ai/v1 and use an sk-mira-… key — the official Python and JavaScript SDKs work unchanged.

The model line

  • mira — the default. A fast model for chat, generation, light coding, classification, and most tool-using agents.
  • mira-thinking — a deeper, slower model that allocates extra compute to step-by-step reasoning. Use when correctness matters more than latency.

The API shape

One canonical endpoint: POST /v1/chat/completions. Send the usual { model, messages }payload, get the OpenAI-shape response back. Streaming is server-sent events, identical to OpenAI's.

Get started

Quickstart

Your first call in under a minute — curl, Python, JavaScript, any OpenAI SDK.

API reference

Base URL, authentication, request and response shape.

Models

When to pick mira vs mira-thinking. Latency, context limits, pricing.

Streaming

Server-sent events, partial messages, and how to consume the wire format.