The stack
layer 01 available now

Every layer, inside one boundary.

Inference is deployed and serving today. The agent runtime, the substrate under it and the deployment layer above are built to land on that same ground — same perimeter, same keys, same audit log. The stack climbs; the boundary doesn't move.

the stack · L01 live
the ground floor is the live one
The three layers[01/03]

Read it from the ground up.

The bottom layer carries production traffic now. Everything above it is marked for exactly where it stands — nothing here is described as finished before it is.

L03
Deployments

Ship an agent the way you ship a service — versioned, started, rolled back — into the boundary that already serves your models.

coming
L02
Agent runtime

Somewhere for the agent itself to live: sessions that last as long as the work, an isolated machine per run, state that survives it.

in development
L01
Private inference

Every open-weights model behind one endpoint inside your perimeter — text, embeddings, image, video, voice. Deployed and serving.

available now
What holds at every layer[02/03]

The guarantees are the constant, not the features.

A platform that climbs is only worth anything if the promises at the bottom survive the climb. These four do not change shape as layers land on top of them.

01

One boundary

Nothing above Layer 01 gets its own network path. Each layer lands in the deployment you already approved — your VPC, your on-prem racks, or dedicated capacity we operate for you.

02

One set of keys

The credentials that reach the inference endpoint are the credentials that reach everything built on it. No second integration, no second key store, no second thing to rotate.

03

One audit log

Every layer writes to the same record: what ran, which model served it, what it touched. A fleet of agents is auditable the same way a single request is.

04

One way out — none

The egress posture does not loosen as the stack climbs. A sandbox, a memory store and a scheduled fleet inherit the perimeter the plane was deployed behind.

Where to start[03/03]

Start at the layer that already runs.

Deployments begin with inference, because that is the layer serving traffic today. The rest arrives on ground you will already be running — and the teams we scope with now are the ones shaping what lands next.

Tell us which layer you need first — we'll scope from thereTalk to us →