How do I use the Plan Review Assistant?
Quick answer: It's the Cursor-style chatbot pinned to the right of your pre-check. Click Assistant top-right to open the drawer. Hover any finding card and click the chat icon (bottom-right of the card) to spawn a conversation scoped to just that finding — the AI loads the sheet image and kicks off an analysis automatically. Or hit + in the drawer for a general conversation indexed over all findings. Conversations live as tabs: open many, rename them, switch between them.
Opening the drawer
On the Findings tab, the Assistant button is top-right (next to the progress ring, Bluebeam, and Finalize & Issue). Click it and the drawer slides in from the right, splitting the screen into three panes: findings list | sheet viewer | assistant.
The button is solid green when the drawer is open; click again to hide.
The two ways to start a conversation
1. Hover a finding → click the chat icon → "focused" conversation
- On the Findings tab, hover any finding card. A chat icon appears at the bottom-right of the card (you'll also see the card's pins brighten on the sheet).
- Click the chat icon.
- A new tab opens in the assistant drawer, auto-titled with the finding's name (e.g., "Electrical Panel Size ..."). The tab badge is a pin number colored by status (red Fail, orange Flag, green Pass, purple Needs Info).
- A green context pill shows what's loaded: something like
pc-17 + Sheet A3 loaded— the finding ID plus the sheet image it lives on. - The AI kicks off an initial analysis automatically. You don't type anything to start.
The response typically includes:
- Sheet vision — the AI quotes what it actually sees on the sheet. For example: "Looking at your sheet A3, I can clearly see the issue in the Electrical Symbols Legend at the bottom right: 'ELECTRICAL PANEL (PANEL SIZE TO BE DETERMINED BY ELECTRICAL CONTRACTOR) 200 AMP'. This is a textbook deferral statement — and city plan checkers will reject it every time."
- What to do — step by step with code citations styled as a reference block (e.g., CEC 230.42, CEC 240.4).
- Related findings to fix at the same time — cross-references to other findings (e.g.,
pc-18 / MEP-ELC-005 — AFCI Protection (FAIL) — your panel schedule must designate AFCI circuits). These are hyperlinks — click them to jump to that finding.
Clicking the same finding's chat icon later jumps back to the existing tab instead of creating a duplicate.
2. Click + at the top of the drawer → "general" conversation
A new tab indexed on every finding in the project. The context pill reads General conversation — all N findings indexed.
Empty state shows four quick actions:
- Summarize critical issues — runs "Summarize all the critical issues in this review"
- Group by who can fix — runs "Group findings by what the architect can fix vs needs consultants"
- Draft corrections response — runs "Draft a corrections response letter for all FLAG findings"
- Biggest rejection risks — runs "Which findings are the biggest risk for rejection?"
Click any and the AI runs it, then auto-renames the tab based on the first message (e.g., "Summarize all the cri...").
The response for Summarize is typically a structured triage: section headers by discipline (Cover Sheet, Structural, MEP, …), Markdown tables with columns like | ID | Issue | Fix Needed | Effort |, and emoji headers (🔴 FAIL FINDINGS, 🟠 FLAG FINDINGS).
Tabs
Every open conversation is a tab at the top of the drawer, Cursor-style:
- + on the far left opens a new general chat.
- Active tab has a white background and subtle shadow; inactive tabs are muted.
- Tab badge on the left is the pin number colored by status for focused chats; for general chats it's a chat-bubble icon.
- Label is auto-named from the finding title, or from the first user message for general chats.
- Hover a tab → × appears on the right — close.
- Double-click the label to rename. Enter or click-away to save, Escape to cancel.
Tabs scroll horizontally if you open a lot.
Message input
- Type
@to open an autocomplete over findings. Each row shows: pin badge (colored by status) + pin number + title + finding ID (e.g.,pc-17). Type to filter (title, category, or ID). Click an item and it inserts@Finding 18: AFCI Protectioninto your message. The AI treats it as loaded context. - Placeholder:
Ask about findings...with a hint@ to reference a findingbelow. - Enter = send. Shift+Enter = newline.
- Responses stream in real time.
What context the AI has
- Focused conversation: the specific finding + the sheet image it's on. The AI has vision on the sheet — it can read title blocks, quote legends, check sheet indexes, cross-reference against other sheets it's seen.
- General conversation: every finding in the project, indexed by text.
- @ mentions inside a message: adds those findings to the working context for the next turn.
What the AI is good at
- Explaining why a finding is a Fail vs. a Flag.
- Citing the code section behind a correction with the code label prominent (
CEC 230.42,CRC R302.1,CBC 1008.4). - Proposing response language you can paste into a corrections letter.
- Spotting cross-sheet inconsistencies (e.g., floor area on CS doesn't match A1).
- Rolling up themes — "show me everything about Title 24" or "every finding on A3".
- Suggesting which findings to fix together in one revision cycle.
What it's not
- It can't change your findings. It reads; it doesn't write to the review. To approve / flag / edit a finding, use the finding card's controls — see Triage findings.
- It can't generate the corrections letter by itself. That's the Finalize & Issue button — see Finalize & Issue.
- It's not the same as the city chatbots (Ask AI in the top nav). Those are open-ended permit assistants for anyone; the Plan Review Assistant is scoped to your specific project. See Ask AI — the three chatbots.
Quick workflow
- Open your pre-check.
- Click Assistant top-right.
- Click + and pick "Summarize critical issues" to get your bearings.
- Scan the findings list. When one doesn't make sense, hover its card and click the chat icon to open a focused tab.
- For each focused tab, read the AI's initial analysis. Ask follow-ups. Use
@to pull in other findings as you need. - Decide: verify, flag (with note), edit, or dismiss — see Triage findings.
- Close tabs you're done with. When the progress ring is full, click Finalize & Issue.