Skip to Content

Documentation

Start here

By product

When it goes wrong

Looking for what a specific tool does? The toolbox lists every one of them.

Start here

Getting started

From nothing installed to a first real answer from the model you have open. Ten minutes — mostly the download.

1. Install

Two routes, same add-in.

From the Autodesk App Store. The App Manager installs it — no copying, no NETLOAD. Start the host and the AEC Pilot tab is there.

From aecpilot.com. One installer finds your Autodesk products and registers with each.

What gets installed

A .NET add-in inside the host and a small bridge executable for AI clients. No Python, no Node.js — and nothing left running after you close the host.

2. Activate your licence

Click About on the AEC Pilot tab and paste the licence key from your email. In Revit the window also opens by itself on first launch.

No key yet? Create an account for a 14-day trial key — no card needed.

Activate once, not eight times

One seat covers all eight applications on the same machine. Activate in one, and the rest follow.

3. Open the chat panel

Click Chat on the AEC Pilot tab. Pilot Chat docks like any palette and works on the open document — nothing to upload or sync.

The Pilot Chat panel docked inside Revit, showing suggested opening questions.

Pilot Chat, docked. The cards on the empty state are live prompts — clicking one sends it.

The first question is the hardest — so click a suggestion and watch it work before writing your own.

4. Ask for something real

Skip “hello”. Ask for something you’d normally click through, so you can judge the answer:

  • “What’s in the active view? Break it down by category.”
  • “Which elements are missing a Mark value?”
  • “List every warning in this model, grouped by type.”

Read-only questions change nothing — the right place to start. How to ask covers phrasing that makes edits land first time.

5. Decide how you want to drive it

Two ways to use the same tools — use both.

The panel inside the application. Nothing else to install. Add your own provider API key in Settings, or use the model already configured.

Your own AI client. Click Connector to start a local MCP server for Claude Desktop, Claude Code, ChatGPT / Codex, Cursor or Antigravity — one click, see Connect your AI client.

The chat panel’s settings: provider and model, response depth, an approval switch for custom code, and a running spend total.

The panel’s own settings. Response depth trades speed for thoroughness; the approval switch makes custom code show you a preview before it runs.

Two switches worth knowing: Response depth trades speed for more multi-step work, and Require approval before custom code previews code before it runs — off by default, worth turning on at first.

Your key, your spend: the panel totals today, this month and all time by provider, and warns you once if you set a daily or monthly limit.

Nothing leaves the machine on its own

The connector listens on localhost only and is off until you start it. Model data is read locally. With your own provider key, only what you send in the chat goes to that provider.

Where to go next