Chat
The chat panel is the fastest way to get answers out of a drawing set. Ask anything in plain language; the agent reads the set, answers with citations, and can run real work for you: checks, comparisons, reports, filled-out forms.
Use case: Ask "which sheets show the electrical room?" and jump straight to them.
Asking questions
Type a question the way you'd ask a colleague:
"Which sheets show the electrical room?"
"What's the fire rating on the corridor walls in Area B?"
"Does the door schedule match the doors tagged on the plans?"
Every answer comes with citations: click one and the viewer jumps to the exact spot on the sheet, so you can verify the answer yourself in one click.
You can also:
- Attach files (images, spreadsheets, PDFs, Word documents, CSVs) for the agent to analyze alongside the set. See Fill in any form.
- Full-screen the chat when you need room for a long thread.
Where the work actually happens
Each chat thread gets its own workspace - an isolated environment the agent works in for as long as the thread lives. That has three consequences worth knowing:
- Long jobs keep going. A multi-step task doesn't have to finish inside one reply. The agent works, reports back, and the workspace is still there for the follow-up.
- Follow-ups pick up where you left off. Step away, come back and ask "now do the same for the mechanical set" - the thread resumes with everything it already worked out, rather than starting cold. After a long gap - more than about six hours - the project opens a fresh composer instead, and the old thread is one click away in the tab strip or the history.
- It can produce real files. Spreadsheets, reports, filled-in forms and tables are written as actual files, returned in the thread, and saved to the project as generated outputs.
You don't manage any of this. It's the reason the chat can be given a job rather than just a question.
Plan, explore, execute
For bigger jobs, the chat doesn't just answer, it works. Give it a multi-step task and it plans the steps, explores the drawings, and executes in one thread:
"Find every fire-rated wall, list the doors on them, and flag any missing a rating."
The agent shows its plan as it goes and cites what it finds. What it can do on your behalf:
| Ask it to | And it will |
|---|---|
| Search the set | Read pages, search text across every sheet, and pull up what it finds |
| Run checks | Dispatch checks from your library across a document or the whole project |
| Write a check | Turn a description or a pasted checklist into runnable checks - see Custom checks |
| Refine a check | Propose an edit to a check's prompt when its findings aren't landing right |
| Flag a problem | Raise an issue with a citation, straight into Issues & Comments |
| Pull existing findings | Read the project's open issues and reason over them |
| Recall earlier work | Search your previous threads on this project rather than re-deriving an answer |
| Set up a view | Open a document or line up an overlay for you to look at |
| Change a markup | Move or reword a callout that's already on the sheet, including issue clouds and their connected text - not just draw new ones |
| Draw a diagram | Make a board - a process, a sequence, a set of relationships - or change one you already have |
| Produce a deliverable | Write a report, build a table, fill in a form, generate a spreadsheet |
If it can't do something, it says so rather than guessing.
Project memory
Tell the chat how your team works once, and it remembers across the whole project. It saves what you tell it as individual facts, in six kinds:
| Kind | What it holds | Example |
|---|---|---|
| Correction | Something the extraction got wrong | "E-101 is electrical, not civil" |
| Preference | How to work this project | "Always metric", "ignore the /Cut Sheets folder" |
| Context | A fact that isn't in the documents | "The east wing is already built and barricaded" |
| Decision | A human ruling, or how documents relate | "The 2026-04 set supersedes the March one" |
| Glossary | A project-specific term | "'WH' means water heater on this job, not wall hydrant" |
| Caveat | A data-quality note about the set | "Sheets A-201 to A-210 are scanned, not vector" |
These are facts you supply, never the agent's own conclusions - its findings live in Issues instead. That separation is deliberate: it means a fresh review is primed by your team's decisions and never by a stale verdict from a previous conversation.
How to use it:
- Save something: just say it. "Remember that we always review to the 2021 IBC on this job." It confirms what it saved.
- See what it knows: ask. "What do you remember about this project?"
- Take something back: say so. "Forget the metric preference, this job is imperial." The old fact is retracted, not deleted, so the record of what changed survives.
Everything active is loaded automatically at the start of every conversation, so you never have to re-brief a new thread. Checks read it too - a check run uses your recorded facts when judging what applies, which is why "this segment is built and barricaded" stops a code check from re-flagging it on every run.
Working across several threads
You're not limited to one conversation per project. Start a new thread for a separate line of work and switch between them from the chat header - useful when a long job is running and you want to ask a quick question without disturbing it.
A long thread carries a lot of context, and past a point the agent starts losing track of earlier steps. The platform tells you when a thread is getting heavy and a fresh one would serve you better - your runs and results are saved either way.
There's also a chat on the home screen, outside any project, for questions that span projects ("which of my jobs have open fire-rating issues?") or for starting work before you've decided which project it belongs to.
Repeatable work
If the chat does something you'll want again, ask it to save the workflow so you can re-run it. See Workflows for saved multi-step reviews, and Custom checks for turning a standard into a runnable check.
When the chat recommends a tool you haven't used, it offers a short "Here's how" video inline. The full clip library is in the Help menu → User Guide.