The measurable half of a quoting engine, as an API.

CAD drawing        customer uploads DWG / DXF / DWF
     ↓
CADLens            parse → entities, layers, geometry, metrics
     ↓
dimensions         lengths, areas, perimeters, counts, extents
     ↓
your logic         material, process, machine time, margin
     ↓
quotation          the number you actually sell

The boundary sits deliberately after measurement and before pricing. Measurement is a solved, verifiable problem that benefits from being shared infrastructure; pricing is where your product differentiates, and it should not live inside somebody else's black box.

• Cut and path length — metrics.length per entity, already computed for lines, polylines and arcs, so you are not re-deriving arc lengths from angles.
• Enclosed area — metrics.area for closed profiles, and sheet-level area and boundingBox for blank size and nesting estimates.
• Feature counts — summary.statistics.byType gives counts per entity type across the drawing; circles usually stand in for holes, and their radius is in the geometry.
• Process routing — sheets[].layers with names and per-layer entity counts, mapped to operations by your own layer conventions.
• Notes and tolerances — TEXT and MTEXT entities carry their string value and position, so callouts can be surfaced to a human reviewer rather than silently ignored.

Open a real result and click through the entity table to see these fields populated on an actual drawing.

There is more than one reasonable way to get geometry out of a CAD file. Which one fits depends on what you are willing to operate, not on which is "better":

• An open-source DXF library (for example ezdxf). Excellent for DXF, in-process, no per-parse cost, and you keep full control. It does not read DWG, so you still need a conversion step in front of it, and layouts, block expansion, hatch boundaries and rendering remain your code.
• A clean-room DWG library (for example LibreDWG). Removes the licensing question and reads DWG directly. Coverage across the long tail of DWG revisions varies, so expect to build fallbacks and a corpus of test drawings.
• The ODA File Converter, self-hosted. This is the industry-standard conversion path — CADLens uses it internally. Running it yourself means a licence, a headless Linux environment for a GUI-era binary, process supervision, timeouts, and a queue in front of it, before you have written any parsing at all.
• Autodesk Platform Services. Deep, first-party, and strong when you want viewing, model derivatives, or the wider Autodesk ecosystem. It is a platform rather than a single parsing call, so integration surface and output shape are correspondingly larger.
• Convert to PNG or PDF. The right answer when a human needs to look at the drawing — and the wrong one when software needs to measure it, because a raster has no layers, no units and no entities.
• Build the whole layer in-house. Sometimes correct, particularly if CAD parsing is your product. Budget for the parts that are not the first parse: format revisions, paper-space layouts, MTEXT formatting codes, extents and unit handling, very large drawings, conversion timeouts, and the on-call that comes with all of it.

CADLens is the hosted abstraction: one endpoint, a versioned output schema, and none of that operated by you. If the trade-offs point elsewhere for your team, they point elsewhere — that is a real engineering decision, not a marketing one.

The pipeline above is generic. Two narrower write-ups go into the specific fields and layer conventions each domain relies on:

• Manufacturing quotation — hole counts, cut length, and part geometry from customer part drawings.
• Laser cutting — profile extraction, closed-path detection, and per-process layer routing.