Nothing changes in your design
until you approve it.

The agent stages the change. You apply it.

The agent can propose, and it can create. Changing engineering state that already exists requires your approval. Creating over a world you already have is not creation, and it is refused — so the only route from an intention to an existing artifact runs through a proposal you accept.

Ask the agent to modify a world directly and it files the change as a reviewable transaction and refuses — APPROVAL_REQUIRED, nothing written, here is what it would have done. The twenty-five capabilities below are sorted by what they are permitted to do, not by what they are called.

Not a log line written afterwards — a record with a lifecycle, kept as it happens. Verified means the artifact was read back and the requested change was proven to have landed.

A plan is computed from facts, and each one carries where it came from — so a number in a proposal can be traced back to whether somebody stated it, the system derived it, the model was measured, or the agent assumed it.

One canvas — code, circuits, and 3D / CAD

A single project surface that morphs between a 2D schematic editor, a 3D scene, and orthographic CAD views — the same parts read as a circuit, a model, and a measured drawing.

Several answers, simulated, side by side

Ask for a change and you do not get one arrangement — you get a set. Each candidate is worked out, simulated, and scored on the same rows, so choosing is reading a table rather than trusting a paragraph. Selecting one turns it into a staged transaction; it still writes nothing until you approve.

“The AI decided” — checkable, turn for turn

A session is a model that asks before it answers: it observes the world, queries what it needs to know, designs, simulates, and comes back with a set. The whole trace is recorded — every turn, every tool call, every fact it cited — and a recorded session can be replayed and compared against the run that made it.

Hub, Agent, Browser, Editor & Terminal — side by side

Every surface is a resizable pane in one window. Open what you need from the toolbar, the Customize-Layout menu, or the command palette — drag the dividers, choose your layout.

It reads the files your industry already writes

No proprietary project format standing between you and your work. The Studio opens what your tools already produce, and runs what your work already needs.

Compose agents from building blocks

An agent isn't a black box — it's assembled from composable pieces in the workspace. Add what it should know, follow, react to, and reach.

The same runtime, from any shell

A session-first agent terminal. Describe a goal in plain language and the agent chains its own tools — read and edit code, search the workspace, run git, fetch the web — to carry it out. Zero dependencies, just Node 18+; run tensoreq from PowerShell, Command Prompt, bash or zsh.

Fifteen commands — each a slash command and an agent tool

Everything the agent can do, you can run yourself by name. Type / for a live, filtered menu; /login with Google or GitHub opens a session.

One project model, nine engineering domains

A project is born into a domain — the Studio gives it that domain's bus, safety boundary and runtime shape. The model spans nine domains; the build is early and honest, starting with Aerial.

Shipped, not promised

Everything on this page is in the build today unless the row below says otherwise.

Talk to the developer

Questions, bugs, partnerships, or hardware / board integrations — reach the TensorEQ developer directly.