Ntersection — AI Gateway

1 · Generate an enrollment token

A GPU box joins the fleet by dialing OUT to this gateway over TLS — no inbound port, no public IP, no manual certificate. The token is one-time and valid 24h.

Node name (permanent identity) Location shown on the fleet card

Tier

Slots

2 · Install on the GPU box

Windows

Linux

a. Install the inference runtime. Install Ollama for Windows and pull the models this box should serve (it listens on 127.0.0.1:11434; the agent is the only thing that talks to the network).

ollama pull nemotron-3.5-lightning:latest

b. Download the agent (PowerShell, from this gateway):

c. Register this box — one command, no browser trip. Because you're already signed in as admin here, this enrolls and approves the box from its own terminal:

Fill in a node name above and click Generate.

Manual alternative (enroll here, approve on the Access tab)

Generate a token above and this fills in.

d. Run as a Windows service (NSSM; run PowerShell as Administrator):

Generate a token above and this fills in.

3 · Approve it

The one-command register in step c already approved this box — it'll appear active on the Fleet tab as soon as it connects, no further action. Only the manual path leaves a node unapproved: approve it on the Access tab. Once approved, the scheduler routes to it immediately — no restarts anywhere. Routing matches by model name; premium-tier nodes are preferred for premium consumers but the fleet never fails closed.

Enrolled nodes

name	state	enrolled

Consumers (API keys)

name	class	node routing	req 30d	tok 30d	today	state

New consumer

People apps-plane profiles — signup is invite-only

username	plan	apps	machines	joined	state

No profiles yet — mint an invite to bring the first person in.

Link a machine to a profile with the owner button on the enrolled-nodes table — that person can then publish apps on it self-serve from the portal.

Teams shared app ownership — created self-serve in the portal

team	plan	members	apps	created	state

No teams yet.

Node leaderboard from the last 200 relayed requests · gateway-observed, never node-claimed

#	node	live tok/s	avg tok/s	best	requests	tokens

No requests observed yet — this fills in as inference flows.

Recent inference requests

when	question	model	node	consumer	tok	tok/s

Nothing yet.

Published apps

app	node	visibility	status	served	share

No apps published yet.

Publish an app

The app must be running on the node's own loopback (e.g. a web UI on 127.0.0.1:3000). It gets https://<slug>.apps.ntersection.ai. unlisted = share-link only; public = anyone, listed on the landing page; private = owner's share links only.

Ntersection / AI Gateway