PDF markup meets AI takeoff — in your browser.
The unified AEC document workspace. Mark up drawings, measure to scale, run AI quantity takeoff, and collaborate live — without bouncing between desktop apps.
Inspector
RFI #042
Callout · Open
AI takeoff
Built for the firms tired of paying $700/seat for a 2003 desktop app
Design partner roster — joining the closed beta in May 2026
The whole pitch
One product, the strengths of two categories.
Legacy markup tools own annotation. AI estimators own takeoff. Vellum is the first product to do both, in a browser, on one canvas, with a single price tag.
| Capability | Markup tools | AI estimators | Vellum |
|---|---|---|---|
| Web-native (no install, any device) | — | ||
| High-fidelity markup tools (13+ shapes) | partial | ||
| Scale-aware measurements (length / area / count) | |||
| Real-time collaboration (live cursors + op-log) | — | partial | |
| AI-powered quantity takeoff | — | ||
| Sheet revisions with pixel + structured diff | partial | — | |
| Per-seat pricing under $50/mo | — | — | |
| Free tier for solo estimators | — | — |
Vellum Studio
Real-time collaboration that feels like Figma — for blueprints.
Op-log architecture means every markup is an independent entity with a vector clock. Two estimators on the same sheet. No merge conflicts. No “last saved at 3:47” panic. Cursors stream over WebSocket; ops persist server-side; reconnect replays cleanly.
- Live cursors with name labels and presence avatars
- Op-log replays missed events on reconnect — no lost markups
- Threaded comments with @-mentions on every markup
- Per-user audit trail (who changed status, when, from where)
Updated RFI #042 — moving the panel from W3 to W2.
@sarah confirmed with electrical sub. Marking resolved.
Photo pin attached on A-201 — panel is in. Looks good.
Rooms
3 pending review
avg confidence
Doors
5 pending review
avg confidence
Windows
2 pending review
avg confidence
Fixtures
9 pending review
avg confidence
19
Pending
87
Accepted
6
Rejected
AI takeoff
Run takeoff in 60 seconds.
Review what matters. Reject the noise.
We send each sheet through Claude Sonnet 4.6 with a structured tool-use prompt. Rooms, doors, windows, fixtures come back with bounding boxes and a calibrated confidence score. Every accept / reject becomes labeled training data for our Phase-2 detection models — the moat compounds with every project you run.
- Detection types: rooms, doors, windows, fixtures (v1)
- Calibrated confidence — never raw model output
- One-click accept turns a detection into a real markup
- p95 latency < 60s per sheet end-to-end
Sheet revisions
“What changed in Rev C?” — answered in 3 seconds.
Upload a new revision; we rasterize both pages at the same zoom, classify every pixel into add / remove / change, and overlay a per-page diff. Existing markups migrate forward automatically — no re-pinning RFIs every revision.
- Pixel diff overlay with red/green/amber classification
- Structured “what changed” list (vector path + text deltas)
- Markup forward-migration with provenance trail
- Side-by-side comparison mode for senior review
60fps
Pan/zoom on 200MB drawing sets
<60s
p95 AI takeoff latency, per sheet
<100ms
Markup fan-out across collaborators
$24/seat
Pro tier — 1/30th of incumbent pricing
Pricing
Per-seat. No surprises. Cancel anytime.
Education and registered nonprofit firms get the Team tier free, forever.
Free
$0
For solo estimators learning the ropes.
- 1 user
- 3 active projects
- AI takeoff: 5 sheets/mo
Pro
$24/seat/mo
The standard plan most firms land on.
- Unlimited projects
- AI takeoff: unlimited
- PDF/XLSX/JSON exports
Team
$49/seat/mo
For multi-office firms with custom workflows.
- SSO + audit log export
- Custom report templates
- Priority support
See the full feature matrix on the pricing page.
Bring a real PDF.
See the difference in five minutes.
Spin up an account, upload a drawing, run AI takeoff, export a markup report. No demo call required.