Bring the shape
SVG and DXF drawings (layers, blocks, splines, units), text in a single-line or outline font, simple shapes and formulas, or meshes for 3D: STL, OBJ, 3MF and glTF.
CAM
From a drawing to a proven program.
Studio computes toolpaths for this machine. Give it a drawing, a mesh, a line of text or a shape your AI describes; it writes the program, sets spindle speed, depth and feed from the twin's physics, and runs it to the end in the twin before you see it.
How it works
Shape in, proven program out.
Every step works in the CAM workspace or through your AI assistant over MCP; both use the same engine on your computer.
Choose the operations
Pockets, contours, slots, holes, text and 3D roughing and finishing, with the tools in the machine's tool table. Studio writes the tool changes and measures each new tool.
Let the twin set the cut
From the material and the tool, the twin picks a spindle speed on a stability lobe and limits each depth by chatter, spindle power, the fixture's grip and tool deflection.
Run it in the twin
By default the program runs to the end on a hidden copy of your setup. Operations that chatter or overload are rewritten and run again; then the feeds are tuned line by line.
Operations
2.5 axes and 3 axes, for this machine.
Programs for the MC-1's controller, with true arcs (G2/G3) wherever the path allows. In the CAM workspace: 2D shapes, seven operations, nesting and several setups; DXF, meshes and 3D and the other operations through the AI assistant.
Pockets with islands
Offset, adaptive (constant engagement, no lifts between regions) or zigzag clearing; a separate floor pass; rest machining with a smaller tool where the first could not reach.
Contours and slots
Inside, outside or on the line, with tabs that hold through-cut parts; trochoidal slots wider than the tool; chamfers and deburring along edges.
Holes and threads
Drilling and peck drilling, helical boring and thread milling, with retract heights checked against what is on the table.
Text and engraving
Single-line text, or outline text in Noto Sans or a font on your computer, pocketed or V-carved along its centre line; where the V-bit runs out of depth, an end mill takes over.
3D roughing and finishing
Layered adaptive roughing, then parallel, waterline, or steep and shallow areas apart. The scallop height is held to what you ask (0.01 mm by default), pencil passes clean the corners, and the shank is checked against the part.
Nesting and two-sided parts
Parts nested on a sheet, small ones inside the holes of large ones. A job can have several setups: the part turns over in place, and each setup is generated and tried in the twin in turn.
Cutting data
Every number has a reason.
Speeds and feeds start from a table of 17 materials and the tool. Then the twin takes over: the spindle speed goes onto a stability lobe, and each depth is cut back until the cut stays stable, under 80 % of the spindle's power, within half of what the fixture can hold and, for finishing, under 0.01 mm of tool deflection. The program says which limit set each value.
- Spindle speed from the stability lobes, not just from the surface speed
- Depth limited by chatter, power, the fixture's grip and deflection
- Feeds tuned line by line after a clean trial run, never into a heavier cut
Proof
Proven in the twin before you see it.
Unless you switch it off, every generated program runs to the end on a hidden copy of your setup, starting from the part as it is now. Studio probes the result for overcuts and leftover material at 0.02 mm, finds collisions and rapids into stock, and says where the tool cannot reach. Operations that chatter or overload are rewritten and run again, up to four rounds.
- Overcut and leftover material checked at 0.02 mm
- Chatter and overload repaired automatically, up to four rounds
- The same input gives the same program, byte for byte, on Windows and Linux
In the twin
What it did on our test parts.
Measured in the twin with nominal machine parameters.
1 requesta 40 mm heart-shaped pocket, 4 mm deep: 143 s of machining, no overcut, no leftover
99.999 %of the isogrid panel (example e14) within 0.02 mm of its design: its shipped program run to the end in the twin, 52 min, no alarm
99.9998 %of a 3D relief within 0.02 mm of the design surface, without overcut
Simulated with nominal parameters until the MC-1 prototypes are calibrated against the twin.
CAM + AI
Your assistant can describe the part.
Over MCP an AI assistant imports a shape with geometry_import and has the CAM write the program with program_generate — then checks, simulates and optimizes it on the twin like any other program. It describes shapes as data: outlines, text, formulas.
- geometry_import keeps the shape with a name and a version, and answers with its size, area, smallest inside radius and a preview
- program_generate answers with the program, the reason for every cutting value and the result of its trial run
- Works with Claude, ChatGPT, Cursor, VS Code or LM Studio, or with a model on your own computer
What the CAM does not do (yet)
So that you know when to reach for another CAM.
- Three axes. A job can have several setups by turning the part over about its long or short edge; Studio does not decide which operation goes on which side, and there is no fourth axis.
- No STEP or other solid formats: export STL or 3MF from your CAD.
- Programs are written for the MC-1's controller; other machines and controllers are not supported yet.
- Cutting data is computed with nominal machine parameters until the prototypes are measured; the reasons in the program say what each value is based on.
- In the CAM workspace itself: 2D shapes (basic shapes, polygons, formulas, SVG or single-line text) and seven operations — pocket, contour, slot, face, drill, bore and engrave — with ISO 286 fits on bores and slots, nesting and setups. DXF, meshes and 3D, V-carving, chamfers, thread milling, reaming, tabs on a single contour, trochoidal slots and outline text go through the AI assistant.
- Very large 3D jobs take minutes to compute: roughing a relief the size of the MC-1's travel took 8 min and finishing it 4 min on two cores of a desktop PC. Past its limits — 8 setups, 5 million triangles, 16 kinds of part in one nest — Studio says so and what to change.
FAQ
Do I still need another CAM?
For three-axis parts on the MC-1, usually not. For four or five axes, turning, STEP models or another machine, use the CAM you know; Studio checks and simulates its G-code like any other program.
Does the CAM cost extra?
The CAM is part of Studio. The free Learning edition has pockets, contours, slots, holes, text and V-carving, and 3D parallel finishing. Maker and Pro add adaptive clearing, every 3D finishing strategy, rest machining, two-sided parts and nesting — see Pricing.
Do I need an AI assistant?
No. The CAM workspace works without AI. An assistant is another way in — for example to describe a part in words.
Does it work offline?
Yes. The CAM runs on your computer. With a local model in LM Studio or Ollama, nothing leaves your computer even when an assistant drives it.
Can I trust the cutting data on my machine?
It is computed for the MC-1 with nominal parameters — the physics model's estimate — and then run in the twin. Each MC-1 is to ship with a twin profile set to its own measurements, so the numbers will follow your machine; on any other machine treat them as a starting point.
Draw it. Prove it. Cut it.
Open the CAM workspace, place a shape on the blank and run it in the twin — the program that comes out is ready for the MC-1.