Skip to content
Noviqent.

AI relay & AI management · A Noviqent product

One gateway for every model and agent you run.

Airelay puts one OpenAI-compatible endpoint and one key in front of 45 LLM providers and 2,376 models — with smart routing, failover, guardrails, RAG, semantic caching and budgets built in — and runs an MCP gateway beside it. Bring your own provider keys and pay their cost plus a flat 4%.

Get started freeFive-minute quickstart

Try Gemini Flash, Groq-hosted Llama and Mistral Small with no provider key and no card — up to 20 requests a day.

app.py — the only change
import osfrom openai import OpenAI client = OpenAI(-    api_key=os.environ["OPENAI_API_KEY"],+    base_url="https://ai.noviqent.co.uk/v1",+    api_key=os.environ["AIRELAY_KEY"],) reply = client.chat.completions.create(    model="anthropic/claude-3-5-sonnet",    # ...or a routing policy: "policy/primary-with-backup"    messages=[{"role": "user", "content": "Hello"}],)
LLM providers behind one API
45
models in the catalog
2,376
MCP servers in the agent directory
15,063
AI plugins you switch on org-wide
12

The platform

Four things you'd otherwise stitch together

An LLM router, an AI firewall, an MCP gateway and an AI cost and usage analytics stack — as one product, one dashboard and one bill.

AI Gateway

One OpenAI-compatible API and one key in front of 45 LLM providers and 2,376 models, with routing that keeps answering when a provider doesn't.

  • Bring your own provider keys — your provider bills you directly
  • Fallback, load-balancing, cost, latency and semantic routing
  • Per-key circuit breaker and a response feedback API
Explore AI Gateway →

AI Guardrails & Plugins

Switch on PII redaction, prompt-injection defence, semantic caching, prompt templates, third-party guardrails, RAG and token budgets for your whole organisation, without touching your app.

  • PII redaction including UK identifiers, with optional restore
  • Semantic cache, prompt compression, LLM-as-judge scoring
  • Token- and cost-based rate limits and budgets
Explore AI Guardrails & Plugins →

Agents & MCP

Connect your AI agents to tools from a directory of 15,063 remote MCP servers, behind one governed MCP gateway and an A2A agent proxy.

  • One-click connect from the MCP directory
  • MCP gateway with API-key or OAuth 2.0 (Noviqent SSO) auth
  • A2A proxy for agent-to-agent calls
Explore Agents & MCP →

Observability & Governance

See every AI request, its tokens and its cost — and control who can change what — with analytics, a request log, team roles, 2FA and an audit log.

  • Requests, p95 latency, errors, tokens, cost and cache hit rate
  • Owner, admin, developer and viewer roles with invitations
  • Audit log of every configuration change
Explore Observability & Governance →

Everything on every plan

Routing, AI plugins, the agent directory and analytics aren't upsells. You pay your provider's price plus 4%.

See pricing →

How it works

Every request, through one policy-driven path

Your apps and agents talk to Airelay. Airelay authenticates them, applies the plugins you've switched on, routes the request — and records what happened.

Requests from your apps and AI agents enter Airelay, which authenticates them, applies guards, rate limits and budgets, serves cache hits and enriches prompts, routes each request, and records analytics — then forwards it to LLM providers using your own keys, or to MCP servers and agents.
  1. 1

    Connect your provider keys

    Add OpenAI, Anthropic, Google, Azure, Bedrock or any other supported provider's key. It's encrypted at rest — or kept in your own Noviqent Vault space — and never shown again.

  2. 2

    Change one line

    Swap your SDK's base_url for Airelay's and pick a model as provider/model — or reference a routing policy as policy/<name>. The rest of your code stays the same.

  3. 3

    Turn on what you need

    Redact PII, block prompt injection, ground answers with RAG, cache answers, cap tokens and spend, or connect MCP servers — from the dashboard, without a redeploy.

Continuous service

Your AI keeps answering — even when one provider doesn't

Reference a routing policy as model: "policy/<name>" and Airelay decides, per request, where it goes.

Fallback chains

Rate-limited, erroring, timing out or returning nothing usable — Airelay moves to the next model in the chain automatically, with no code change on your side.

Smart load balancing

Weighted, lowest-latency, lowest-cost, lowest-usage, priority tiers — or semantic routing that sends each prompt to the model best described for it.

A fast circuit breaker

A provider key that keeps failing is skipped for a cooldown instead of every request waiting through a slow timeout — and your app can report bad answers via the feedback API.
See every routing strategy →

45 providers, 2,376 models, one API

Connect a key and call the models that provider hosts — each one can be a fallback, a load-balance entry or a semantic route for another. Providers include:

  • OpenAI
  • Anthropic
  • Google Gemini
  • Google Vertex AI
  • Azure OpenAI
  • AWS Bedrock
  • Mistral
  • Cohere
  • Groq
  • Together AI
  • DeepSeek
  • xAI
  • Perplexity
  • Hugging Face
  • IBM watsonx
  • Cerebras
  • SambaNova
  • Fireworks AI
  • DeepInfra
  • AI21
  • Nvidia NIM
  • Databricks
  • Cloudflare Workers AI
  • Replicate

Trust

Your keys, your data, UK-hosted

Airelay was built for teams who need to know exactly where their prompts and credentials go.

Bring your own keys

Your provider bills your own account at their own price. Airelay never resells tokens and never pools your traffic onto a shared key.

Keys never sit in the gateway engine

Provider keys are stored encrypted, or in your organisation's own Noviqent Vault space, and injected per request — not written into the routing engine's own config.

UK-hosted, residency-aware

The gateway runs on Noviqent's UK-hosted infrastructure. Pin Azure OpenAI, AWS Bedrock, IBM watsonx.ai, Databricks, Alibaba Cloud Model Studio keys to an EU/UK region; providers without region pinning are labelled as such.
Security & data residency in detail →

Put one gateway in front of all of it

Free to sign up, no card needed to try the free models. Connect your own provider keys whenever you're ready.

Get started freeRead the quickstart