Detumble.ai - 2026 Pricing, Features, Reviews & Alternatives

Detumble.ai is a cloud-based web application hosted on AWS (EU region, Stockholm) that targets quantity surveyors, estimators, BIM teams, structural engineers, and contractors working on reinforced concrete and steel building projects. The platform ingests AutoCAD DWG or DXF files (or a zip of either), reads every sheet, schedule, note, and bar schedule, and produces a coordinated set of structural deliverables: a material take-off, a tender-coded bill of quantities, a 3D IFC4 model, a knowledge graph, and an RFI register. All outputs remain drafts until a human engineer accepts them at a review gate, preserving the engineer of record's responsibility throughout.

Detumble.ai combines large language models (supplied by Anthropic via API) with deterministic arithmetic to process drawing sets. The AI component reads extracted drawing text and geometry; deterministic software then places elements and computes quantities from the confirmed reading. No geometry is placed by the model alone, and no customer content is used to train AI models without explicit written consent. Every quantity is traceable to the sheet and schedule row it came from, and conflicts are surfaced as queries rather than silently resolved. The platform also exposes API and MCP endpoints for teams that need to integrate structural take-off into wider workflows. Plans cover individual estimators (Solo, Studio, Tender tiers) and enterprise team accounts for multi-team firms.

From a single confirmed reading of the drawing set, Detumble.ai produces multiple coordinated deliverables simultaneously: a line-by-line material take-off covering concrete, rebar, formwork, and excavation; a BoQ mapped to tender codes and ready to price; an IFC4 model with each structural element traced to its source sheet and schedule row and exportable to Revit or any BIM viewer; and an RFI register that packages the highest-priority assumptions as formal RFIs sorted by tender priority. An in-app 3D viewer with per-element evidence and a chat interface for querying the confirmed reading support ongoing project analysis. Each run is quoted in credits before it starts, and failed runs return the full credit reservation.