Airfy Nexus

The central knot. Open source. On your GPU.

Nexus, Latin for connection. The central knot where every thread comes together autonomously. The open-source AI engine that ties Talk, Chat, Metis, and your network into one operator-controlled reasoning loop. Runs on the GPU you own.

Open source
Hardware-portable
Standalone-first
nexus / src / config.ts
MIT
1// nexus.config.ts, the integration knot
2import { Nexus, Talk, Chat, Metis } from "@airfy/nexus";
 
4const nexus = new Nexus({
5 gpu: "local", // your Blackwell, your boundary
6 model: "open://llama-3.1-70b", // any open model, any quant
7 agents: [Talk, Chat, Metis], // first-class participants
8 network: airfyNetwork ?? null, // optional, not required
9});
 
11// ask the system anything; it routes through every Airfy app
12await nexus.ask("what changed in our Austin Factory last night?");
airfy/nexus
Self-hostable
Pin any commit

The three things Nexus refuses to be.

Not closed source. Not silicon-locked. Not Airfy-Network-required. Nexus is the engine the rest of the AI Platform plugs into, and the engine you can replace if we ever stop deserving the trust.

Auditable.

Every line of Nexus is readable. Pin to a commit. Build from source. Replace the model. The kind of openness regulators ask for and lawyers can sign.

Hardware-portable.

Nexus runs on Nvidia Blackwell, Hopper, or whatever your operator policy allows. Pluggable inference, no silicon lock-in. Replace the engine without touching the apps.

Standalone-first.

Nexus can run without Airfy Network. The integration is a power-up, not a prerequisite. The day you adopt Airfy Network underneath, Nexus inherits the richer signal.

Where the threads meet

Five inputs. One loop. One source of truth.

Airfy Talk
Voice → reasoning loop
Nexus
Airfy Chat
Conversation → reasoning loop
Nexus
Airfy Metis
Knowledge graph → reasoning loop
Nexus
Airfy Network
Topology + telemetry → reasoning loop (optional)
Nexus
Nexus
Surface the answer to the operator
Atlas
The integration story

Talk to your company. Chat with your system.

Voice your question. Chat your idea. Query your knowledge. Every answer runs on the GPU you own, not on a vendor's cloud. The five Airfy AI products turn your operator network into a nervous system you can speak to.

Airfy Talk
Voice your question. The network listens.
Airfy Chat
Team chat on your GPU. Slack-shaped, operator-owned.
Airfy Metis
Wisdom layer built from Talk + Chat.
All on
Your GPU
Nvidia Blackwell-class GPU server running every Airfy AI App
Your hardware, your boundary. Nothing leaves your rack.

Stop being training fodder for someone else's model. When you use Slack, your conversations train somebody else. When you use Airfy Chat on your own GPU, they do not.

See the AI Platform

Buy it the way that fits the room you are in.

Airfy products are sold through partners who handle the install, the trust, and the local ownership. Operator-scale buyers can still talk to us directly.

Through your ISP partner

Bundle Airfy with your broadband line.

Subscribers get the Home App and the operator runs Compass on the AI-Network Controller. Same product, your provider's name on the bill.

Through your AI consultancy or MSP

Let an Airfy partner deliver and run it.

MSPs and AI consultancies install the network, set up the AI products on your GPU, and run the AI App Store with you. Local trust, Airfy stack.

Operator-scale direct

Buy directly when the deal is operator-scale.

For ISPs, sovereign customers, or enterprise rollouts above 100 sites: talk to us directly. We connect you with the right partner for the install if you prefer.

Bring the GPU. We bring the open source.