Skip to content
AI,  BIM

Connecting Project Files to an AEC AI Assistant: Access Checklist

By ArchiLabs

Date Published

You can connect project files to an AEC AI assistant by preparing a small, approved package, confirming what the tool can actually open under your user account, and starting with read‑only tasks before allowing a supervised edit in a copy. The assistant helps most with intake, documentation, and tightly bounded model changes; your team owns access choices and acceptance checks.

Choose project files for an AEC AI assistant pilot

Start small so you can see real work without broad exposure.

• Files to include
• One Revit model copy (two if your task depends on a link), with links relocated into the pilot folder.
• Optional: 1–2 DWGs that reflect common drafting details.
• 4–8 PDFs (a few sheets and details) that show naming, title blocks, and conventions.
• One small spreadsheet, e.g., a sheet register or equipment list.
• Context the assistant can use
• A 1–3 minute screen recording narrating the task goals.
• 3–5 annotated screenshots pointing to example views/sheets.
• A “Standards” subfolder with a sample title block, view templates, tag examples, and schedule filter notes.
• Sanitize carefully
• Remove client names and PII from filenames and visible sheet text.
• Inspect embedded metadata: Revit project information, title block parameters, DWG xrefs, PDF properties, and any hyperlinks in families or schedules. Scrub or replace as needed.

Important: some desktop tools act through the signed-in user’s account and may have its file access; confirm the actual architecture and permissions for the tool you use. Folder names or polite prompts do not isolate access. Verify what the tool can open from your machine/account before proceeding.

Check file access and open the project in a controlled workflow

Copies can still point back to live links or a central model. Follow your documented worksharing and copy procedures:

• Have the BIM lead prepare the test model using the project’s documented worksharing/copy procedure, including detachment where appropriate. Verify its relationship to central before editing.
• Repath all links to the pilot copies using relative paths where possible; confirm none reference production locations.
• Use read‑only open modes where supported by your storage and tools; only move to an editable test copy if your task demands it.
• When you do allow edits, work in a clearly named copy (for example, “Working_[Model].rvt”) that is separate from any production file.

Integration availability and behaviors depend on your application version, plan, accounts, and permissions, and disconnecting a tool does not remove content you’ve already shared—manage any uploads where you uploaded them.

What an AI assistant can help extract from project files

For this pilot style, AI contributes in three places:

• Intake and scoping: From your screen recording, annotated screenshots, PDFs, and spreadsheets, it can assemble a to‑do list, point out missing inputs, and propose a minimal file checklist aligned with your examples.
• Documentation: It can create and update native views, sheets, tags, and exports in supported applications when you provide templates and naming rules. A copied or duplicated Revit view is not a frozen geometry/parameter snapshot; treat it as live data and review accordingly.
• Targeted model actions: For a bounded scope, it can create/update native model elements and route systems under supervision.

If you plan in‑model work, review the current integrations for your environment on ArchiLabs Desktop. Do not attribute browser‑workspace features to Desktop.

Test file reading, documentation, and one approved model edit

Use a three‑step sequence that starts read‑only and ends with a supervised edit.

1) Read‑only inventory and briefing

• Ask the assistant to list views and sheets it can see, highlight unresolved links, and summarize naming patterns from your examples.
• Checks: verify link paths are inside the pilot folder; spot‑check 3–5 sheet names against your convention. Row counts in a schedule do not equal element counts unless you have itemization and filters set.

2) Documentation task

• Provide rules for one new view and one new sheet (templates, naming patterns, title block). Request an updated sheet index as a spreadsheet or PDF, as supported in your setup.
• Checks: confirm the new view/sheet follow your naming and templates; if you export a schedule, specify its filters and itemization explicitly—schedule exports do not automatically carry your model settings. A schedule of selected objects does not prove other objects stayed unchanged.

3) Limited model change (copy only)

• In “Working_[Model].rvt,” request a small, well‑bounded change (e.g., place a small number of symbols or update text on a specific sheet). Avoid edits that would require overwriting From Room/To Room, Design Option, or other parameters that are not ordinary spreadsheet fields.
• Checks: verify geometry and parameters by selecting the placed/edited elements. Do not infer engineering correctness from a warning count or a system color. When changing label text or title‑block fields, confirm whether that parameter is type‑based; do not modify a shared type to affect selected instances without an approved type strategy.

Project access example: compare a sheet register, then edit a description

This example is hypothetical.

Suppose you assemble “AI‑Pilot_OfficeL3_2026‑11” with AR_OfficeL3.rvt (detached copy), 6 record PDFs, Sheet_Index.xlsx, 4 annotated screenshots, and view template files in “Standards.”

• Task A: Ask the assistant to compare Sheet_Index.xlsx with sheets in AR_OfficeL3.rvt and propose adds/removals to align the register. Provide one screenshot of a correctly named sheet as the pattern.
• What you review: the proposed list, the naming rule application, and any sheets the tool could not find by number/title. Remember: a schedule export won’t include filters unless you specify them, and a row total isn’t proof of completeness without itemization set.
• Task B: Approve creation of one sheet (for example, “A‑L3‑GEN‑101”), applying your title block and placing an existing floor plan view. Request an updated Sheet_Index.xlsx reflecting the change.
• What you review: that the sheet uses your title block and naming; that the view placed is the intended one. A duplicated plan view is still live; confirm view template and scope box are correct.
• Task C: In Working_AR_OfficeL3.rvt only, ask for a minor description text update on that new sheet (for example, change a sheet‑level Description parameter from “Office Level 3 Plan” to “Office L3 – General Plan”).
• What you review: whether “Description” is an instance parameter on the sheet. If it is type‑based in your title block family, do not change it globally; instead, use an approved instance‑level field or a separate type per your standards.

This example is illustrative; it avoids MEP routing or multi‑file edits and keeps validation human and specific.

An access and file-handling checklist for AEC AI workflows

• Confirm which account and permissions the assistant uses and which locations are available through the chosen connection.
• Models are detached from central (if applicable) and saved in the pilot folder; all links are repathed locally.
• Start read‑only; enable edits only in “Working_[Model].rvt” when required.
• Standards provided: naming rules, view templates, title block, schedule filters/itemization notes.
• Sanitization reviewed, including embedded metadata and hyperlinks.
• Acceptance relies on element/parameter inspection and visual comparison—not only counts, exports, colors, or warning totals.

For scope realism and guardrails, see What AI Can and Can’t Do in Revit Today and share essential context using the patterns in Revit Model Context Protocol. When you’re ready for in‑model tasks on Windows with native application integrations, confirm availability for your setup on ArchiLabs Desktop.

AI Agents for Architecture
AI

Learn about how Model Context Protocol (MCP) is changing the way we use tools like Autodesk Revit and the future of AI agents and architecture