One OpenAI-compatible endpoint. The right model, per prompt.

aiix.io is an LLM routing gateway at api.aiix.io with a per-prompt routing brain: a small classifier inspects each request — task type, complexity, tool needs — and dispatches it to the highest-scoring model on our eval leaderboard. Cost and provider health are tie-breakers, not drivers. Set model="aiix/auto" and stop guessing which model wins your workload.

Start free — bring your own keys   Talk to us about enterprise

How the routing works

Per-prompt dispatch

Every request is classified before it leaves us: task type, complexity, whether it needs tools, rough context size. Target: under 150ms of overhead, and a classifier failure falls back to your policy default — routing never fails your request.

Eval-driven leaderboard

A scheduled eval worker runs test suites against every enabled pool model and writes scores per task type. Candidates are the models that clear the quality bar for your task; the top score wins.

Failover built in

On a 429, 5xx, or timeout we retry the next same-tier candidate once, before the first streamed byte. A per-provider circuit breaker (open at >20% errors in 30s) feeds the tie-breaker, so traffic steers away from unhealthy providers on its own.

OpenAI-compatible /v1/chat/completions, streaming included. Native adapters for OpenAI, Anthropic, Google, Mistral, and Bedrock, plus a generic OpenAI-compatible adapter for any BYO endpoint (vLLM, llama.cpp, DeepSeek, Groq) — with SSRF-guarded base URLs.

Pricing

Free — bring your own keys

$0
  • Self-service signup with email verification
  • Your own provider API keys, stored KMS-encrypted — you pay the provider directly
  • Full routing brain: aiix/auto, default-model control, leaderboard selection
  • Rate-limited per org (token bucket) — this tier exists to prove the router works

Sign up free

Enterprise

Custom, Stripe-billed
  • Routing policies and default-model control per org
  • Audit feed: metadata-only event journal (org, model, tokens, latency, status)
  • Budgets and usage caps; SSO on the roadmap
  • Sold on trust: consent state is a contractual commitment, not a checkbox

Contact enterprise@aiix.io

Your prompts are yours — the co-op, in plain terms

We process every request to route it — that is the product. What we commit to is persistence: prompt and completion content is journaled only when your org opts in (an org-level toggle, with a per-request header override), and the consent flag is stored with every record. Non-consented content is never persisted, not even encrypted.

The data co-op — our monetization line — sells prompt/completion datasets sourced exclusively from that explicit opt-in partition. Consented data is PII-scrubbed nightly into a separate S3 bucket before it becomes a dataset; the scrub pipeline is the only path into it, and access is control-plane-gated and audited. Co-op content is never resold without consent, and any sale requires legal review first. Turn the toggle off and no further prompt content from your traffic is stored.

Point your client at aiix/auto

Signup takes a minute: create an org, verify your email, add provider keys, get an sk-aiix-… API key. No credit card on the free tier.

Create your account