TOOLCAD
Agentic text-to-CAD framework with an interactive CAD modeling gym that rolls out tool-augmented trajectories against a CAD engine with hybrid feedback, built to train and compare open-source LLMs as CAD tool-using agents through online curriculum reinforcement learning.
What it does and how it connects
TOOLCAD is an agentic text-to-CAD framework that augments large language models with programmatic, CAD-specific tools and trains them end to end via reinforcement learning for text-to-CAD generation. The model invokes operations such as sketch constraint solving, feature assembly and parametric history editing, and produces editable, manufacturing-aware B-Rep models directly from natural-language descriptions.
An interactive CAD modeling gym rolls out tool-augmented trajectories against a CAD engine with hybrid feedback and human supervision, and an end-to-end post-training strategy elicits a CAD modeling chain of thought and refines the agent through online curriculum reinforcement learning. The paper reports that reinforcement learning on tool-aware trajectories substantially improves geometric validity, parametric editability and user-intent alignment over prompt-only or end-to-end supervised baselines, while remaining compatible with any open- or closed-source LLM backbone.
The project page is the only public artifact found at research time: it carries the abstract, method and highlights, with the results gallery and quantitative table marked coming soon, and no code repository was found, so no repo, repo stats or license are recorded for this page.
Alternatives
Compare side by side- CADEngBench
A two-track benchmark of parametric parts and assemblies, rather than a framework that trains text-to-CAD agents with reinforcement learning.
- Engine Bench 0.1
Two models race head to head to build one engine, rather than a framework that trains text-to-CAD agents with reinforcement learning.
- cad-silent-failure-bench
Measures silent failures where CAD code runs but the part is wrong, rather than a framework that trains text-to-CAD agents.
- Embodied CAD
Solver-grounded agents that iteratively build parametric B-Rep assemblies, rather than a framework that trains text-to-CAD agents with reinforcement learning.
- text-to-cad-mcp
An MCP server that turns text into CAD models, rather than a framework that trains text-to-CAD agents with reinforcement learning.
Sources
Not verified
- Pricing: none stated on the project page or paper page.
- Real user voices: no real linked post read for this page; this is a paper with a project page and no code repository found.
- License: no license is recorded; the catalog license of source-available is unverified because no code repository was found. The project repository gongyifeiisme/toolcad-project contains only the website.
- No public code repo was found, so no repo stats are recorded.
Last verified November 6, 2026. Listings are informational. The evidence score measures public, verifiable evidence and is not an endorsement or a compatibility claim.