Airelay
One gateway for your AI providers, your own APIs, and your AI agents - with guardrails, caching, analytics and failover built in.
Airelay is a gateway platform with three jobs. It puts one OpenAI-compatible endpoint in front of 45 AI providers and 2,376 models, so your code changes a base URL instead of an integration. It publishes your own APIs through the same engine, with authentication, rate limiting, caching and transformations you configure rather than build. And it fronts AI agents and MCP servers, so a tool your team connects is reached through one audited, authenticated door.
Bring your own provider keys - each billed directly by that provider, never pooled - and pay a flat 4% on top, metered per request. Keys are encrypted at rest or held in your own Noviqent Vault space and resolved only at the moment a request needs them; they are never stored inside the gateway engine itself. A curated free tier works with no key connected at all.
Routing is a policy, not a code change: fallback chains, weighted load balancing, lowest-latency, lowest-cost, lowest-usage, priority tiers, or semantic routing that picks a model from the meaning of the prompt. A per-key circuit breaker skips a provider that has failed repeatedly instead of every request rediscovering the same outage.
Fourteen AI policies run inside the gateway on every matching request, with no application changes: PII redaction (including UK identifiers such as NI numbers, NHS numbers, postcodes and sort codes, optionally restored in the reply), semantic caching, prompt and response guards, third-party guardrails (AWS Bedrock, Azure AI Content Safety, Google Model Armor, Lakera, or your own endpoint), retrieval from your uploaded documents, prompt compression, token and cost budgets, and automatic response-quality scoring.
Everything is visible: requests, errors, p95 latency, tokens, cost and cache hit rate, with a searchable request log, an audit trail of every change, team roles, and a developer portal that publishes your API documentation. Airelay runs on Noviqent's UK-hosted infrastructure.
Features
- 45 AI providers and 2,376 models through one OpenAI-compatible endpoint - OpenAI, Anthropic, Google, Azure OpenAI, AWS Bedrock, Mistral, Cohere, Groq, OpenRouter, Together AI, DeepSeek, xAI, Databricks, watsonx and more, plus your own self-hosted models
- Seven routing strategies - fallback chains, weighted load balancing, lowest-latency, lowest-cost, lowest-usage, priority tiers, and semantic routing by prompt meaning
- A per-key circuit breaker that skips a repeatedly-failing provider for a cooldown instead of waiting through every timeout
- PII redaction with UK identifiers - NI number, NHS number, postcode, sort code, phone - optionally restored in the reply
- Semantic caching: a repeated question is answered from cache instead of being paid for again
- Guardrails from AWS Bedrock, Azure AI Content Safety, Google Model Armor, Lakera, or your own endpoint, plus prompt and response guards by meaning
- Retrieval from your own uploaded documents, injected into prompts automatically
- Token and cost budgets per organisation, key, route or model, with prompt compression to cut spend
- Automatic response-quality scoring on a sample of traffic
- A full API gateway for your own services - routes, consumers, API keys, JWT, OAuth2, ACLs, consumer groups, load-balanced upstreams with health checks, certificates and vault-backed secrets, and 48 policies in total
- An agent gateway for MCP servers and agent-to-agent tools, with a directory of 15,063 MCP servers and 418 agents to connect from
- Analytics, a searchable request log, an audit trail, team roles, and a developer portal for your API documentation
- Bring your own provider keys - encrypted at rest or held in your own Noviqent Vault space, never stored in the gateway engine
- A flat 4% on your own provider cost, metered per request - plus a curated free tier that needs no key at all
- UK-hosted, with per-provider data-residency labelling and optional two-factor authentication
Pricing
Free to start - a curated set of models work with no provider key or card. Pay-as-you-go is a flat 4% on top of your own connected provider's cost, metered per request. Larger API gateway and agent deployments are priced on request.