\n\n
![\"OpenTakeoff](\"docs/img/social-card.png\")
\n\n\n\n![\"A](\"docs/img/mcp-live-demo.gif\")
\n\n*A real run (3× speed): the agent takes off patient rooms 161–163 on a federal finish plan,\nexporting after each commit. Every shape lands in the app as a dashed **pencil proposal** and\nbecomes ink only when the operator clicks Accept.* The full run, live and uncut, is\n[on YouTube (2:47)](https://youtu.be/e--kXxSGv7Y).\n\n### The tools\n\n| Group | Tools |\n|---|---|\n| **Open and orient** | `load_plan` · `sheet_info` · `sheet_context` · `get_sheet_vectors` · `read_sheet_text` · `find_text` · `view_sheet` |\n| **Scale** | `set_scale` |\n| **Measure** | `one_click` · `detect_rooms` · `measure_polygon` · `cut_out` · `measure_line` · `measure_surface` · `place_count` |\n| **Repeat and derive** | `symbol_sweep` · `sweep_schedule_row` · `derive_base` · `derive_transitions` · `apply_rules` |\n| **Read the drawing set** | `sheet_graph` · `resolve_tag` · `find_schedule` |\n| **Edit and audit** | `list_shapes` · `edit_shape` · `edit_condition` · `edit_materials` · `duplicate_condition` · `split_condition` · `delete_shape` · `undo_last` |\n| **Mark and sign** | `annotate` · `list_annotations` · `link_annotation` · `mark_verdict` · `delete_verdict` |\n| **Ask** | `create_rfi` · `list_rfis` · `resolve_rfi` · `delete_rfi` |\n| **Hand off** | `takeoff_summary` · `export_takeoff` · `export_report` · `export_marked_pdf` · `export_dxf` · `import_takeoff` |\n\nPlus browsable sheet resources (`takeoff://sheets`) so an agent can *see* the working set, not\nonly act on it. Multi-document sessions are first-class: a bid set is plans **plus** schedule\n**plus** addenda, and `load_plan --merge` adds a document without disturbing existing scales,\nconditions, or shapes—the sheet graph then spans the whole set, so a room tag on one file\nresolves to a schedule row in another. `edit_condition` reaches the waste %, the ×N multiplier,\nand `roll_setup`, so an agent's takeoff doesn't ship with net === gross.\n\n**The agent's manual is [`docs/AGENT_GUIDE.md`](docs/AGENT_GUIDE.md)**—the counterpart to the\nestimator's: the operating model in six facts, the standard finish every takeoff ends with, the\nwithheld-is-the-answer doctrine, what has no agent verb and why, and a refusal-to-next-move table.\nTool-by-tool reference: [`mcp/README.md`](mcp/README.md). The same surface in prose, with the\nsheet-graph and sweep behavior in depth: [`docs/MCP.md`](docs/MCP.md).\n\n### The contract that makes it drivable\n\nMost measurement APIs are hostile to an agent because they let it be confidently wrong. These\nare the rules that make this one safe to hand a model, and why each one exists:\n\n1. **One coordinate frame, stated everywhere.** Image pixels at render scale 2.0—PDF points\n × 2, origin top-left, y down, the browser canvas's native space. Every sheet payload carries\n dims in both px and pt. No tool takes a coordinate in units it has to guess.\n2. **Scale is a gate, not a default.** The drawn scale note is *read* off the sheet but never\n applied silently; adopting it is always an explicit `set_scale`. Measuring an unscaled sheet\n refuses. Pixels × a wrong scale² is every number wrong at once, so the engine would rather\n stop than guess. Disagreeing scale notes inside a measured region raise a warning rather\n than a silent pick.\n3. **The engine traces; the model doesn't invent.** `one_click` returns the ring the wall\n network produced from a seed point you name. A model cannot hand back a polygon it imagined and have\n it counted.\n4. **Every record carries how it was made.** Method, seed point, whether hatch filtering\n engaged, whether it came off scan pixels, confidence factors,\n and the machine's original ring if a human later moves it.\n5. **Agent work is pencil until a person inks it.** Exports land in the canvas as dashed\n proposals. `mark_verdict` lets an agent sign its own work as a graphite `AGENT` diamond; the\n green `APPROVED` seal has exactly one code path and it is the toolbar button under a human\n hand. No MCP call, no import, mints one.\n6. **The deliverable is a marked-up planset, not JSON.** `export_marked_pdf` burns the work\n into the drawings as drawn—condition colors, hatches, quantity chips, count markers—behind\n a legend cover with totals and a tally of how much of the set a person has actually\n reviewed. A takeoff nobody can check is not a takeoff.\n7. **Refusals are actionable strings.** *\"That space isn't enclosed on the plan linework—the\n fill spilled\"* tells a model what to do next. A silent zero doesn't. Tools that can't answer\n withhold with a stated reason rather than returning a plausible number.\n\n### Prove it — OpenTakeoff Academy\n\n[**aec.kentucky-ai.com**](https://aec.kentucky-ai.com) is a standalone open benchmark and\ncertification arena for agents that do takeoff. Bring any model and your own harness; you are\nscored on **operating a real takeoff tool** against geometry you don't control—a wrong\ncalibration yields a wrong area—not on emitting a plausible-looking number. Runs emit a\nsigned bundle with full provenance of every tool call, scoring is against held-out ground truth\nand a human Senior Estimator baseline, and clearing a tier earns a credential that's\nindependently verifiable. The Certified path drives this engine (`opentakeoff-mcp`) behind the\ntask tools. Repo:\n[Kentucky-ai/opentakeoff-academy](https://github.com/Kentucky-ai/opentakeoff-academy).\n\n---\n\n## For the person at the canvas\n\nThe agent path exists because the human path is real. Everything below is the production\nmeasuring engine carved out of a commercial Division 9 estimating system—not a demo\nreimplementation.\n\n```bash\ncd web\nnpm install\nnpm run dev # http://localhost:5173\n```\n\nOr open the [**live demo**](https://opentakeoff.kentucky-ai.com). Drag in\n`demo/sample-plan.pdf`, accept the detected scale, choose a condition, press **`A`** (Area)\nand click the room's corners. (One-Click Area is temporarily gated — see the note at the top.) Open **Report** for the breakdown and the exports. That whole loop on\nvideo: [walkthrough (1:10)](https://youtu.be/aHiW8H2TSBs) ·\n[One-Click Area (0:51)](https://youtu.be/YIjWZ-BAhLE). The complete\nzero-to-exported walkthrough is the [**user manual**](docs/USER_GUIDE.md).\n\n**What a real bid looks like on it**, in the order an estimator works one:\n\n1. Drag in the whole `.zip` off the bid platform—plans, finish schedule, addenda.\n2. Set the scale on every sheet you'll measure, and **check a dimension** (`K`) on each. Ten\n seconds a sheet, and it's the only mistake that gets every number at once.\n3. Pull your conditions off the architect's finish schedule instead of typing them, and set waste\n and materials *before* you trace.\n4. Stitch anything split at a match line, align it, and only then start measuring.\n5. **Trace** the floors room by room — Area (`A`) while One-Click is gated. Derive base and transitions off the rooms you just\n traced rather than measuring them a second time—and read what the derivation *reports and\n never counts*, because those are doorway thresholds you still owe.\n6. Walk the set and look at what landed, fix with the grips, save a **revision**.\n7. Export both: the Report for pricing, the **Marked set** PDF for whoever has to check you.\n\nThe full version of that sequence, with the section for each step, is the manual's\n[working order](docs/USER_GUIDE.md#the-working-order-on-a-real-bid). What every term above means\nis in its [glossary](docs/USER_GUIDE.md#18-glossary--what-the-words-mean-here).\n\n### Open anything, instantly\nA plan **PDF**, an **image** (scan, screenshot, photo), or a whole **`.zip` plan set** straight\noff a bid platform. Zips are unpacked and images wrapped to PDF *in your browser*—multi-page,\nmulti-file, up to **4 sheets side-by-side**, with hostile-archive guards so a malformed zip\nfails cleanly instead of ballooning the tab. No upload step, no conversion service, no account.\n\n### A real measuring engine\n**One-Click Area** is the headline — **temporarily gated** while the flood engine is re-validated (see the note at the top); this is what it does when it is on: click inside a room, the linework bounds a flood fill, the\npolygon traces itself, the vertices snap to true corners. **Hatching and poché don't fool it**—tile\ngrids, plank lines, and section fills classify as pattern rather than wall, and the\nescalation is conservative enough that a misread can never come out worse than the strict fill.\n**Scanned sheets work too**: with no vector linework the engine reads rendered pixels—adaptive\nthresholding, polarity detection for blueprint negatives, a gap-bridging pass for\nfaded ink—and badges the result so you verify the edges before committing. On CAD exports\nthat publish a layer tree, One-Click reads the declared roles instead of inferring them.\n\nPlus the full manual kit—**Area, Rectangle, Linear, Curved Line, Surface Area (walls),\nCount**, and **Cut Out** deducts—and a **Zone check** that answers \"what's in this wing?\"\nwithout touching the takeoff.\n\n**⟂ Transitions** derives the line where two finishes meet, from rooms you already measured.\nFinishes changing inside one open space commit as a dashed butt-joint run you accept; rooms\nparallel across a wall are **reported and never counted**, because that transition is a\nthreshold in a doorway no trace can locate—you get its length, the wall thickness, and a link\nthat puts it on screen.\n\n![\"One-Click](\"docs/img/one-click-area.gif\")
\n![\"OpenTakeoff](\"docs/img/report.png\")
\n