Skip to content

Tips for success

Practical habits that make VitruAI faster, cheaper, and more accurate — and the gotchas worth knowing before they bite.

Use the AI efficiently (context matters)

VitruAI is an AI system: it works best when each task has a clean, focused context. A few habits go a long way — they improve answer quality and reduce the credits a check consumes.

  • Start a fresh thread for a new check. Don't continue an unrelated check in a long-running conversation. A long thread carries all its prior history into every turn — that's slower, costs more, and dilutes the model's focus. New task → new chat.
  • Let the Coordinator delegate. For multi-discipline work, ask the Coordinator for the whole sweep and let it spawn specialists. Each specialist gets its own fresh context, so the work parallelizes instead of piling into one bloated thread.
  • Don't paste or upload large documents into chat. A big code PDF or spec dumped into a message bloats context on every subsequent turn. Instead, put durable reference material and rules in your workspace — agents read it there, only when relevant, without carrying it in the conversation.
  • Be specific. Name the rule, discipline, or elements ("check rule 1.5.2", "verify egress on level 2") rather than "check everything" when you only need one thing. Tighter scope = faster, more precise, cheaper.
  • Reuse rules instead of re-describing them. If you check the same requirement often, have the Setup agent author it as a rule once. Running a stored rule beats re-explaining the requirement in prose every time.

Get accurate results

  • Scope work to a project. Verification reports are saved durably and become searchable/exportable only when the check runs inside a project. Without one, you get a chat answer but no saved report.
  • Make sure the evidence is in the model. Rules that verify visually (dimensions, tags, annotations, markings) can only conclude when those elements are present in the relevant views or sheets. If they're missing, expect Indeterminate or NeedInfo — add them and re-run.
  • Indeterminate / NeedInfo aren't failures. They mean the check couldn't conclude, by design, rather than risk a false pass/fail. Provide the missing info or evidence and re-run.
  • Recheck after you fix things. Use Recheck on a saved report to re-verify just the previously flagged items after you've changed the model.
  • Keep your rules and constitution current. Agents read your workspace on every run, so refining the constitution or a specialist's brief immediately sharpens results.

Connection & sessions

  • Keep Revit open with the model loaded, and launch VitruAI from the VitruAI ribbon tab — the agent reads the model through that live connection.
  • Multiple models open? Pick the right Revit session in the session picker so a check targets the model you mean. See Working with more than one Revit session.

Keep an eye on credits

AI checks draw down your firm's shared credit pool. Scoping checks tightly and keeping threads focused (above) directly reduces spend. Admins can watch usage on the Billing page (30-day breakdown by model and member) and top up before the pool runs dry.

Known gotchas

  • SmartScreen may warn about the installer even though it's signed — use the official download and More info → Run anyway.
  • The add-in didn't appear? Install with Revit fully closed; supported versions are 2024 / 2025 / 2026.
  • A few rules are still maturing — if a result looks wrong, send it to support with the report.
  • Multiple firms on one email? Pick the right firm at sign-in; each firm is a separate account.

Full fixes are in Troubleshooting & FAQ.