MillBill agent pairs language models with geometric feedback for machining
Researchers describe an agent that splits machining planning from geometric execution, letting a locally hosted open model refine cutting plans through simulated feedback.
In plain words
MillBill is a system that helps turn a 3D design into instructions a milling machine can follow. Instead of making a language model write every line of machine code, it has the model choose machining steps while separate software checks the geometry and reports what went wrong. The authors say this division improved results on their test parts.
What was announced
The paper's authors present MillBill as an agent that keeps machining decisions apart from their geometric execution and checking. A locally hosted, mid-sized open model receives a boundary-representation part, stock, and material, and proposes operations and cutting parameters for indexed three-axis machining. A separate geometric harness then builds toolpaths, simulates how material is removed, and sends feedback the model can use to revise its plan.
The authors report two notable design and evaluation points. Setup planning considers what the available cutters can physically reach, not just visibility, and the resulting programs are scored by the share of target surface area that lands within the specified tolerance after simulated machining. Using Qwen3.8-27B, they state that MillBill beat frontier-model approaches in which the model itself constructs geometry and toolpath workflows on the evaluated tasks. Code is available on GitHub.
What it does not show
- Evidence is limited to the paper's abstract, so the full methodology, task set, and comparison details cannot be verified from the supplied excerpt.
- Performance claims come from the authors' own simulated evaluations and have no independent confirmation in the sources.
For security and IT teams
- Scope
- The system targets indexed three-axis machining and was evaluated only on the authors' selected machining tasks, per the abstract.
Sources
Source-support review completed 19 Nov 2026. This story was drafted with AI assistance and checked against retained source passages. Automated checking is not independent verification.
Coverage: arXiv: abstract.
Updates and corrections
No corrections so far.
First published 19 Nov 2026
New to the subject? AI from the Ground Up starts from the beginning.