Alayra Nexus · open-source AI gateway

In development

One endpoint for every model. Your keys.

Pool and load-balance provider keys behind one OpenAI- and Anthropic-compatible endpoint — with tiered failover, team budgets, BYOK isolation and a live cost console. Self-hosted, Apache-2.0, and never a cent of markup.

Star on GitHub Self-host in one command

v1.6.5 latest
4 stars
Apache-2.0
TypeScript · Node 22

Benchmarks

Latency and throughput results, with the method to reproduce them, are being prepared for the Nexus site.

Coming soon

How requests are routed

Four rules. No surprises.

Routing in Nexus is deliberate and explainable — the console prints these same rules next to your pools.

01
Tier order
Pools are tried best-first — premium, standard, fast — dropping a tier only when the one above has no key with headroom.
02
Within a tier
Models are ordered by priority, then by price according to your cost weight. Cost is a tiebreaker; it never overrides health.
03
Session affinity
A continuing conversation sticks to the key that last served it, so the provider’s prompt cache keeps paying off.
04
Team isolation
A team’s own keys are tried first. The shared pool is used only if the team allows it — otherwise it stays hard-isolated.

The console

A console your finance team can read.

Everything in one browser UI served by the gateway itself — no separate web server, and it works air-gapped.

Alayra Nexus analytics: requests, success rate, latency, cost and response-cache savings over 90 days

Reliability, speed, spend and savings — success rate, latency percentiles, response-cache savings and per-provider breakdowns.

What’s inside

Production-hardened, end to end.

Routing

Tiered failover: premium → standard → fast
Cache-aware sticky sessions
Optional cost-aware tiebreaking
Model registry with per-1M pricing

Reliability

Per-key circuit breaker, half-open probe
Separate, flat handling for 429s
Fail-closed when Redis is unreachable
Encrypted, portable backup & restore

Cost control

Per-request USD cost by team
Budgets: block, notify or downgrade
Exact-match response cache, $0 hits
Analytics with CSV export

Security

AES-256-GCM keys at rest
Owner, admin and viewer roles
TOTP 2FA, SSO (OIDC), lockout
SSRF protection on by default

Observability

Prometheus /metrics
Optional OpenTelemetry tracing
Health and readiness endpoints
Webhook and email alerts

Developer experience

OpenAI /v1/chat/completions
Anthropic /v1/messages
Embeddings, images and audio routes
Playground: compare models & strategies

Quick start

Running in a minute. Scaling when you are.

Start standalone on a laptop with SQLite and in-process memory — nothing disabled. Move to Postgres and Redis for production; encrypted backups carry every key across.

1# One command. No clone, no Docker, no Postgres, no Redis.
2npx @alayrasystems/nexus
3
4# → full dashboard on http://127.0.0.1:3000
5# → point your app at http://127.0.0.1:3000/v1

Then point your app at it

1from openai import OpenAI
2
3client = OpenAI(base_url="https://<your-nexus>/v1", api_key="<team-key>")
4
5resp = client.chat.completions.create(
6    model="alayra-nexus-1",  # auto-route, or name any model you've configured
7    messages=[{"role": "user", "content": "Hello"}],
8)
9print(resp.choices[0].message.content)

Cursor routes requests through its own servers, so it needs a publicly reachable HTTPS URL; local tools such as Cline, Continue.dev and Claude Code work against localhost.

Nexus Cloud

In development

The same gateway. None of the servers.

Nexus Cloud is the hosted control plane for secure multi-provider AI: your application keeps one secure path to every model, requests go out on your keys, and we run the infrastructure.

Workspaces for every team
Authenticator-based two-factor and recovery codes from day one
Session tokens in HttpOnly cookies — never in browser storage
Bring your own provider keys

Join the Cloud early access

Nexus Cloud workspace portal: welcome screen with security setup and workspaces

How it compares

An honest side-by-side with LiteLLM.

LiteLLM is the closest well-known project. It wins on reach and ecosystem; Nexus wins on what you get without paying, and on the depth of the operator console.

	Alayra Nexus	LiteLLM
Providers	9 presets + any OpenAI-compatible endpoint	100+ built in
Admin dashboard	Built in — analytics, teams, pools, audit, health	Built in
Team budgets	Block, notify or downgrade at the cap	Yes
Key pooling & failover	Per-provider pools, tiered failover, circuit breaker	Load balancing, retries, fallbacks
Audit trail	Append-only, per action, no delete endpoint	Logging integrations
SSO	Included, Apache-2.0	Commercial licence
Two-factor auth	Included	—
CLI & SDK	CLI coming soon · SDK not yet	Mature CLI · Python SDK
Licence	Apache-2.0, every feature included	Open source + commercial tier

Compared against LiteLLM’s public documentation, July 2026, as published in the Nexus README. Spot something out of date? Open an issue — we would rather fix it than win on a stale fact.

FAQ

Questions, answered

Does Alayra add a markup to my AI usage?

No. Nexus sends every request on your own provider keys, so each provider bills you directly at its own price. Nexus computes the cost of every request so you can see it — it never resells tokens.

Is Alayra Nexus really open source?

Yes — Apache-2.0, with every feature included: SSO, two-factor auth, team budgets and the audit trail are in the open-source licence. The Alayra Nexus name and logo are trademarks.

Which providers does it support?

Nexus ships presets for Anthropic, OpenAI, Google Gemini, Groq, Mistral, OpenRouter, Hugging Face, Cloudflare Workers AI and Cerebras, and any OpenAI-compatible endpoint works as a first-class pool. Azure OpenAI, Bedrock and Vertex are reachable through a custom pool today.

Will my existing code and tools work?

Nexus speaks the OpenAI Chat Completions API and the Anthropic Messages API, with streaming and tools. Change the base URL in your SDK, or point Claude Code, Cline, Continue.dev or Cursor at it.

Does Nexus send any data to Alayra?

No. A self-hosted gateway makes zero outbound calls to Alayra Systems or any third party other than the providers you configure.

What is the status of Nexus Cloud?

Nexus Cloud — the hosted gateway — is in development. Join early access and we will reach out as it opens.

Early access

Every Alayra product is in active development. Join early access for Nexus Cloud, Kinetic IDE and Kinetic AIDE — or self-host Nexus today.

Get early access hello@alayrasystems.com