renue Updates Drawing SaaS "Drawing Agent" with AI Agent-Driven Drawing Interpretation Tool Selection and 2D Drawing Auto-Generation Foundation
renue has released a functional update for its drawing SaaS "Drawing Agent," which automatically generates 3D models from 2D drawing images. The update introduces a mechanism where an AI agent selects necessary tools for drawing interpretation and implements a foundation for automatic generation of 2D drawings from minimal design information.
📋 Article Processing Timeline
- 📰 Published: May 1, 2026 at 21:00
- 🔍 Collected: May 1, 2026 at 12:31
- 🤖 AI Analyzed: May 1, 2026 at 13:07 (35 min after Collected)
renue Co., Ltd. (Headquarters: Minato-ku, Tokyo; Representative: Yusuke Yamamoto) has implemented a functional update for its drawing SaaS "Drawing Agent," which automatically generates 3D models simply by uploading 2D drawing images. In addition to conventional processing, a mechanism has been introduced where an AI agent selects the necessary tools to proceed with drawing interpretation. Functions of 3D modeling software such as Rhinoceros have been added to the in-house developed drawing interpretation process and toolified, with the agent combining and calling them for each drawing. Simultaneously, a foundation for automatically generating 2D drawings from minimal design information has been established, enabling design and 3D conversion to proceed from drawings even in areas with few free drawings.
Executive Summary
This functional update adds an AI agent-driven tool selection mechanism to Drawing Agent's drawing interpretation process. Each process required for drawing interpretation, shape extraction, and 3D conversion has been toolified, and the agent determines the calling order based on the input drawing. This configuration allows for adaptation to new drawing patterns without rewriting scripts.
In addition, multiple functional enhancements are being pursued, including the toolification of 3D modeling software functions, a 2D drawing auto-generation foundation, user feedback integration, interactive editing UI, and a self-improvement loop. This update aims to resolve traditional constraints such as lack of flexibility and securing drawing data, one by one.
What is renue's Drawing SaaS "Drawing Agent"?
We provide the drawing SaaS "Drawing Agent," which automatically generates 3D models simply by uploading 2D drawing images.
Service details here
Even without CAD software operation skills, designers themselves can convert 2D drawings into 3D data in minutes. This transforms the conventional conversion work, which used to take CAD operators several hours, into an experience completed simply by uploading a file. Recently, a pre-reference function for part information has been added to "Drawing Agent."
Recently, a "Drawing Cleanup" function utilizing GPT-image-2 has been implemented in "Drawing Agent."
https://prtimes.jp/main/html/rd/p/000000034.000091210.html
Premise of the Target Domain
Towards the automation of drawing interpretation and quantity surveying in manufacturing and construction industries, methods combining AI-OCR and image recognition are widely being considered. However, drawings handled on-site vary in line types, legends, and symbols from company to company, limiting the scope of fixed rule-based processing. The structure requiring tuning for each drawing has been a barrier to automation.
In manufacturing and construction sites, CAD native PDFs, scanned PDFs, and paper drawings coexist. In advancing the automation of drawing interpretation, in addition to the diversity of notation, the characteristics of shapes handled also differ by domain. In areas where experience has compensated, such as product drawings that frequently use curved surfaces, the difficulty of directly converting images to 3D is higher than in other industries.
The accuracy of large language models in tool calling has improved to a level suitable for practical application. An environment is now in place to construct processes similar to human drawing interpretation through a combination of multi-step inference and external tool calls. renue is advancing Drawing Agent's functional updates in response to this change.
Goals
In conjunction with this update, we are proceeding with the implementation of user feedback integration, interactive editing functions, and a self-improvement loop.
Responding to User Feedback
We will establish a mechanism to incorporate user corrections and feedback regarding interpretation results and 3D conversion results as judgment material for the agent. Correction details will be recorded per project and reflected in subsequent processing. This aims for a configuration where individual tuning for each drawing does not require manual work by personnel, and the product evolves in line with on-site operations.
Interactive Boundary and Arrow Editing UI
We will develop editing functions that allow users to immediately correct outline boundaries and direction arrows detected by the drawing interpretation agent directly on the UI. This configuration allows users to quickly intervene in highly difficult judgment processes, aiming to achieve both full automation and practical processing time on-site. The design will ensure that correction results are reflected in downstream 3D conversion processes, and correction history will also be utilized as learning data for the self-improvement loop.
Self-Improvement Loop
We will build a self-improvement loop that feeds accumulated correction history and processing results back into the agent's judgment logic and toolset. This establishes a flow where data obtained from operations directly leads to the expansion of supported drawing patterns and improvement in interpretation accuracy. The aim is to evolve the product into a form where accuracy improves with operation, without relying on manual updates.
Executive Summary
This functional update adds an AI agent-driven tool selection mechanism to Drawing Agent's drawing interpretation process. Each process required for drawing interpretation, shape extraction, and 3D conversion has been toolified, and the agent determines the calling order based on the input drawing. This configuration allows for adaptation to new drawing patterns without rewriting scripts.
In addition, multiple functional enhancements are being pursued, including the toolification of 3D modeling software functions, a 2D drawing auto-generation foundation, user feedback integration, interactive editing UI, and a self-improvement loop. This update aims to resolve traditional constraints such as lack of flexibility and securing drawing data, one by one.
What is renue's Drawing SaaS "Drawing Agent"?
We provide the drawing SaaS "Drawing Agent," which automatically generates 3D models simply by uploading 2D drawing images.
Service details here
Even without CAD software operation skills, designers themselves can convert 2D drawings into 3D data in minutes. This transforms the conventional conversion work, which used to take CAD operators several hours, into an experience completed simply by uploading a file. Recently, a pre-reference function for part information has been added to "Drawing Agent."
Recently, a "Drawing Cleanup" function utilizing GPT-image-2 has been implemented in "Drawing Agent."
https://prtimes.jp/main/html/rd/p/000000034.000091210.html
Premise of the Target Domain
Towards the automation of drawing interpretation and quantity surveying in manufacturing and construction industries, methods combining AI-OCR and image recognition are widely being considered. However, drawings handled on-site vary in line types, legends, and symbols from company to company, limiting the scope of fixed rule-based processing. The structure requiring tuning for each drawing has been a barrier to automation.
In manufacturing and construction sites, CAD native PDFs, scanned PDFs, and paper drawings coexist. In advancing the automation of drawing interpretation, in addition to the diversity of notation, the characteristics of shapes handled also differ by domain. In areas where experience has compensated, such as product drawings that frequently use curved surfaces, the difficulty of directly converting images to 3D is higher than in other industries.
The accuracy of large language models in tool calling has improved to a level suitable for practical application. An environment is now in place to construct processes similar to human drawing interpretation through a combination of multi-step inference and external tool calls. renue is advancing Drawing Agent's functional updates in response to this change.
Goals
In conjunction with this update, we are proceeding with the implementation of user feedback integration, interactive editing functions, and a self-improvement loop.
Responding to User Feedback
We will establish a mechanism to incorporate user corrections and feedback regarding interpretation results and 3D conversion results as judgment material for the agent. Correction details will be recorded per project and reflected in subsequent processing. This aims for a configuration where individual tuning for each drawing does not require manual work by personnel, and the product evolves in line with on-site operations.
Interactive Boundary and Arrow Editing UI
We will develop editing functions that allow users to immediately correct outline boundaries and direction arrows detected by the drawing interpretation agent directly on the UI. This configuration allows users to quickly intervene in highly difficult judgment processes, aiming to achieve both full automation and practical processing time on-site. The design will ensure that correction results are reflected in downstream 3D conversion processes, and correction history will also be utilized as learning data for the self-improvement loop.
Self-Improvement Loop
We will build a self-improvement loop that feeds accumulated correction history and processing results back into the agent's judgment logic and toolset. This establishes a flow where data obtained from operations directly leads to the expansion of supported drawing patterns and improvement in interpretation accuracy. The aim is to evolve the product into a form where accuracy improves with operation, without relying on manual updates.