Changelog

What changed, newest first. The API is versioned by the /v1 path prefix.

Versioning policy

ℹ Note

Changes within /v1 are additive. We may add new optional request fields, response fields and models without notice. We will not remove or repurpose an existing field, or change a default in a breaking way, within /v1. A breaking change would ship under a new version prefix. The model catalog itself changes continuously and is not part of the version contract — always read it from GET /v1/models rather than hardcoding ids.

2026-06-21 — v1.0 · Initial release

The first public version of the AIx inference API.

API

OpenAI-compatible surface: /v1/chat/completions, /v1/embeddings, /v1/images/generations, /v1/audio/speech, /v1/audio/transcriptions, /v1/models.
Legacy native endpoints retained (/chatcompletions, /image-generation, /text-to-speech, /audio-translation, /embeddings).
Bearer-token and X-API-Key authentication.
OpenAI-style error envelope across every endpoint.
Streaming chat via SSE (chat.completion.chunk).
Advanced passthrough on OpenAI-compatible providers: tools / tool_choice, response_format (JSON mode / schema), and multimodal (vision) content.
Downloadable OpenAPI 3.1 spec.

Platform

Multi-provider routing with priority/cost ordering, automatic failover and a Redis-backed circuit breaker.
Prepaid credits (1 credit = $1) with estimate-hold → settle-on-real-usage billing and a reconciliation queue.
Scoped API keys: per-key allowed models/kinds, daily/monthly spend caps and TPM limits.
Per-key rate limiting (100 requests/minute).
Idempotency via the Idempotency-Key header.
Zero prompt retention; keys hashed at rest.
Razorpay and PayPal top-ups.
Prometheus /metrics endpoint and a public status page.

Docs

Full guide set (migrating from OpenAI, coding agent, tool calling, structured outputs, vision, streaming UI, editor setup), concept deep-dives, and a complete API reference.

FAQ →