Concepts¶
A short tour of how VitruAI works and the terms you'll see across the app, the add-in, and your reports.
The agent workflow¶
VitruAI verifies a Revit model with a team of AI agents:
- QAQC Coordinator — the orchestrator you talk to. It scopes your request, decides which rules and disciplines to run, and dispatches specialists. It never runs rules itself; it coordinates and assembles the report.
- Discipline specialists — per-discipline agents (e.g. Structural, MEP) that the Coordinator spawns in parallel via Orbit. Each reads the locally installed Revit instance (elements, parameters, rooms, geometry, views, sheets) and runs its discipline's rules.
The flow:
- You ask the QAQC Coordinator to verify the model.
- The Coordinator scopes the request and the relevant disciplines.
- It spawns discipline specialists in parallel.
- Each specialist reads the model and runs its rules.
- Findings flow back to the Coordinator.
- The Coordinator assembles a decisive verification report.
When a check needs visual evidence, VitruAI can also use sheet images in addition to raw model data. To go deeper on agents, see Agents, tooling & delegation.
Rules¶
A rule is a checkable requirement — a project requirement, design rule, or regulatory/code check (for example, a minimum room area). You can state rules in the chat, or run against rules already configured for your project. Each rule produces a result for the model and, where relevant, for individual elements.
Findings, statuses, and severity¶
A check produces findings. Each rule (and each element under it) carries a status:
| Status | Meaning |
|---|---|
| Passed | The rule is satisfied. |
| Failed | The rule is not satisfied. |
| Indeterminate | The check couldn't conclusively determine pass/fail. |
| NeedInfo | More information is required to complete the check. |
Findings also carry a severity — one of Critical, Major, Minor, or Info — and, where available, the affected element IDs and a suggested fix.
Why some rules are Indeterminate or NeedInfo
Some rules can only be verified visually, and only when the required annotations, dimensions, tags, or markings are present in the Revit views or sheets. Others need extra information to be fully verified. These surface as Indeterminate or NeedInfo rather than a false Passed/Failed.
Reports — chat summary vs. the full report¶
There are two views of a check's results, by design:
- In chat you always get a concise summary of the findings, plus a reference to the saved report — the chat stays readable and doesn't dump every element.
- The full, durable report — every rule, every checked element, statistics, and suggested fixes — lives in the Reports tab of the Revit add-in, and can be exported to PDF or Excel.
Reports are saved per project
A verification report is persisted when the work is scoped to a project. Without a project in scope, the QAQC Coordinator still answers in chat, but there's no durable saved report to open later. See Projects & access.
The web app does not have its own reports page — the Reports tab in the add-in is where saved reports live.
Firms, members, and roles¶
VitruAI is multi-tenant: your account belongs to a firm. Within a firm, each member has a role — owner, admin, or member — that governs what they can manage (team, billing, projects, devices). Access to individual projects is granted per member as viewer, contributor, or maintainer. See the Firm-admin guide for the full model.
Connected devices & sessions¶
The Revit add-in connects to your firm account as a device, paired with a per-device token. You can have several devices connected at once — multiple machines, or multiple Revit sessions — and when more than one is live, you pick which Revit session a check targets from a session picker. Commands route by that session (not just your account), so two open models never collide. Admins can see and revoke devices — see Connected devices.
The credit-pool & seat model¶
VitruAI billing has two parts:
- Seats — each active member is a seat (billed per seat per month).
- A prepaid credit pool — your firm's shared balance that AI verification work draws down. Some plans also include a per-cycle credit grant, and members can have a per-cycle usage cap.
When the pool is empty (or a member hits their cap), checks are blocked until the pool is topped up or the cap resets. The full mechanics — plans, top-ups, caps, and the Stripe billing portal — are in the Firm-admin guide.
Beyond QA checks¶
Inside the web app you'll also find:
- Workspace — your firm's git-backed file tree of rules, the QA constitution, and agent briefs, with a built-in viewer. It's the source of truth agents read when they check a model.
- Orbit — the delegation layer: the Coordinator spawns discipline specialists that run in parallel, each shown as a live card with its own transcript and a timeline of events.
- Authoring checks — write, compile, and validate your own rules.