Nestbolt

简体中文 · 日本語

In the frozen v1 benchmark, 8.5% less material at the median than the best of Deepnest, SVGnest and Sparrow at equal 60-second budgets, and 9.6% at 300 seconds, in the final benchmark run (fresh seeds, ten held-out jobs). Sparrow still wins 8 of the 56 jobs at 60 seconds (details, every number).

Free, open-source nesting for people who cut fixed sheets: laser cutters, CNC routers, plasma, waterjet, fabric and leather. Nestbolt arranges parts so they use the least material. It runs in your browser (nothing is uploaded), on the command line, and as an MCP server that AI agents can call.

Open the browser app · Download a release · Connect an AI agent

How Nestbolt compares with other ways an AI agent can nest parts (OpenNest, pyFit, Sparrow, Deepnest, SVGnest and more), by interface, file formats and sheet model: nesting for AI agents.

Try the v0.2.0-alpha.1 preview for editable LightBurn project nesting, lead-in allowances and improvements for large curved drawings. Windows x64, Linux x64 and macOS Apple silicon tools are available, plus the Windows MCP bundle. Their published-download checks pass. Import acceptance and common-line material use remain under development; the v1.1 plan is 60% complete. Stable v0.1.0 remains available.

The hosted browser app runs a preview build from main (newer than v0.2.0-alpha.1). The benchmark figures below belong to the frozen 75cbb32 build; this preview makes no new comparison against competing tools.

451 parts on one 850 × 425 mm sheet in 120 seconds on one CPU core: DejaVu Sans Bold glyphs in four sizes with their exact Bézier outlines, 90° rotations and 1 mm spacing. 29 letters sit inside the holes of others, and the unused material stays as one clean offcut along the top. The layout passed the independent validator. Rebuild it with python scripts/make_showcase.py --nest 120.

How it compares

Final acceptance run (2026-10-13): 75 frozen sheet jobs (35 academic, 40 maker, ten of them held out from all tuning), equal time budgets, one CPU core per tool, five fresh seeds each, every layout checked by an independent validator. The comparison is against the best of Deepnest, SVGnest and Sparrow on each job, each with the better of its two settings.

| Budget | Valid Nestbolt layouts | Less material than the best competitor (median) | Wins or ties | | --- | ---: | ---: | ---: | | 60 s | 375/375 | 8.5% | 48 of 56 | | 300 s (long subset) | 60/60 | 9.6% | 11 of 11 |

At 60 s Nestbolt uses less material than Deepnest on all 55 jobs where Deepnest produced a layout (median 9.9%) and than SVGnest on all 36 (14.4%). Sparrow, which Nestbolt builds on, wins 8 jobs. Seven are one-sheet jobs that Nestbolt hands to Sparrow itself after its own first layout, so Sparrow runs with slightly less time there; it is 0.03-2.4% better on them. The eighth (two sheets) is 0.34% better. That misses the project's own 60-second bar (wins or ties on 90% of jobs and never more than 1% worse); the 300-second, strip-parity and speed bars are met (status).

One benchmark job exactly as nested in a 60-second development run (m-mechanical-frame-inserts, seed 0, sheet turned to landscape): 138 parts, all 39 inserts placed inside frame windows, everything packed into 56% of a 2440 × 1220 mm sheet so the rest stays one offcut. The best competitor used 62% of the sheet on this job.

In 30 s Nestbolt matches the competitors' 300 s result on 10 of 11 jobs. The first final run (build 484ce8b) stopped with an error on one job; the fix is in 75cbb32 and both runs are published. Every run, including failures, is in BENCHMARKS.md; status lists the evidence per gate. The results page shows every per-job number, tool revision, seed and the hardware; python scripts/reproduce.py --phase final --final-acceptance rebuilds the datasets, reruns what is missing and regenerates it.

What works today

• Import SVG and DXF: lines, arcs, ellipses, Béziers, B-splines, blocks, transforms and units, converted to millimetres. Each closed outline is a part; outlines inside it are its holes. Choose cutting layers, identify stock outlines, and review repairs and unsupported geometry. See import options and the shop-file verification.
• LightBurn projects (.lbrn2 and legacy .lbrn): nest groups as rigid units and export an editable .lbrn2 per sheet, preserving original artwork, text, images and layer settings. Tool and output-disabled shapes stay fixed. Older text may need a separate native save to embed its outlines; mixed active/fixed groups need regrouping before nesting. Import messages explain cached text, bounding rectangles and missing curve controls. See LightBurn preparation.
• Nest on fixed sheets: margins, part and edge spacing, several sheets, several stock sizes with limits, defects, remnant outlines, rotations (free, stepped or a list such as 0/180 for fabric grain) and mirroring. Small parts can sit inside the holes of larger parts. The last sheet is compressed towards one end so the offcut stays usable.
• Export one SVG and one DXF per sheet from the original curves, transformed rather than re-polygonised. LightBurn projects use preserving .lbrn2 output. Export fidelity checks record the tested files and tolerances.
• Lead-in allowances reserve room for leads and set contour starts, with shortened leads and unresolved contours reported explicitly. Red LEADIN paths are review guides; the cutting program creates the actual leads. Imported LightBurn projects keep their editable contents and receive separate SVG/DXF cutting files. See lead settings and CAM setup.
• Common-line cutting, optional and still under development: identify shared straight cuts with kerf compensation. It can reduce cutting distance while increasing material use; two development jobs exceed the planned material limit. The export needs CAM preparation to avoid cutting shared sections twice. See common-line setup and limitations.
• Check every result with an independent Clipper2 validator (overlap, spacing, containment, rotations, quantities). Invalid layouts are never returned.
• Browser app (web/): the same engine compiled to WebAssembly in Web Workers, automatic worker selection, a live view of the best layout, projects saved in the browser or as a ZIP, and five one-click examples. Files never leave the computer. An optional Quick initial layout setting helps start large jobs. Current browser checks and native comparisons are tracked in Status.
• MCP server with import_parts, nest, validate_layout, export_layout and estimate_sheets, tested with Claude Code. Its budgets are repeatable: same build, inputs, work budget and seed give the same layout unless the real-time limit is reached, which is reported explicitly. See the agent interface.

The engine builds on jagua-rs and sparrow by Jeroen Gardeyn and collaborators (arXiv 2509.13329). Fixed-sheet search lays the last sheets end to end as one walled strip and shrinks the last sheet with a port of Sparrow's separator; parts move between sheets and an emptied sheet disappears. Near-rectangular parts such as box panels also get a bounding- rectangle packing start. One sheet of hole-free parts without defects is exactly Sparrow's own strip-packing problem, where it is the state of the art, so there Nestbolt runs unmodified Sparrow for the rest of the budget.

Quick start

Use the browser app without installing anything, or download native tools for Windows x64, Linux x64 or macOS Apple silicon from the v0.2.0-alpha.1 preview or stable v0.1.0. Each archive contains nestbolt, nestbolt-mcp and licences; download checksums are provided alongside it. The separate Windows MCPB bundle installs the server in Claude Desktop; see agent setup.

To build the development version, install Rust through rustup (the repository selects Rust 1.98.1):

cargo build --release -p nestbolt-cli -p nestbolt-mcp

Import parts, nest them for 30 seconds, and get per-sheet SVG/DXF files (examples/box-99x63x45.svg is the browser app's example box):

target/release/nestbolt import examples/parts.svg --output library.json
target/release/nestbolt nest --library library.json --request examples/nest-request.json --out-dir run --seconds 30

For drawings with larger gaps, import --close-gaps 0.05 explicitly repairs up to 0.05 mm and reports every repair's position and size. It is off by default; the silent join limit remains 0.005 mm. The UI and MCP expose the same option. See import options and reports. Lead reservations and contour starts are available in the browser, request JSON and nest --lead-in --kerf ...; see settings and LightBurn setup.

A request names library parts with quantities and the sheets in millimetres:

{
  "parts": [{"id": "parts-1", "quantity": 4}, {"id": "parts-2", "quantity": 12, "rotations": {"kind": "free"}}],
  "sheets": [{"width_mm": 600, "height_mm": 400, "margins_mm": {"left": 5, "right": 5, "bottom": 5, "top": 5}}],
  "spacing_mm": 1.0,
  "edge_spacing_mm": 1.0
}

run/ then holds layout.json, report.json and sheet-0.svg/sheet-0.dxf and so on. Existing files are never overwritten. nestbolt export re-exports a layout; nestbolt validate checks a polygon job and layout. Add --repeatable to nestbolt nest or nestbolt sheets for a deterministic work budget instead of wall-clock time: the same inputs, budget and seed then give the same layout. On a machine much slower than the reference PC, --time-limit <seconds> allows more real time for the same work (the default is --seconds). The optional --quick-start setting tries a conservative initial arrangement before improving it. It is off by default and may use more material. The browser calls it Quick initial layout under Advanced; request JSON and MCP use "quick_start": true. Original geometry, spacing, stock limits and rotation permissions still apply. Unsupported stock or angle rules use the ordinary constructor. File formats are described by the JSON Schemas.

Browser app

Open the hosted app and pick an example (finger-jointed boxes, letters inside letters, a 138-part mechanical kit, a FreeSewing T-shirt