Naver D2SF Makes New Investment in On-Device AI Startup TraceWeave
TECHWORLD ·
✦ AI Summary
Naver D2SF announced on the 21st that it had made a new investment in TraceWeave, an on-device AI startup.
TraceWeave is developing an AI architecture called 'Phase-TITAN' that can understand long context and perform multi-step tasks while keeping memory usage constant.
TraceWeave has built a demo that runs a multi-step agent using only a smartphone CPU, and has completed a paid proof of concept (PoC) in mobility while carrying out performance validation of an in-vehicle AI agent.
Naver D2SF announced on the 21st that it had made a new investment in TraceWeave, an on-device AI startup. Naver D2SF decided to make the investment, saying that TraceWeave has the technical capabilities to implement AI agents even in constrained device environments.
The background to that judgment lies in the limitations of existing transformer-based models. As context length increases, the size of the KV-Cache in existing transformer-based models expands, requiring more memory. However, on-device environments such as smartphones, cars, and robots face major constraints in memory, battery life, and heat, making it difficult to handle long context and to implement AI agents that perform multi-step tasks.
TraceWeave is developing an AI architecture called 'Phase-TITAN' for such on-device environments. 'Phase-TITAN' can understand long context while keeping memory usage constant, and can perform multi-step tasks.
To reduce the memory issues that arise when processing long information, TraceWeave has its own self-developed architecture, 'Phase-TITAN,' designed to ease memory bottlenecks. Phase-TITAN is a structure that replaces the existing transformer architecture, and it is designed to keep memory usage constant even as the length of processed information increases. Based on this, TraceWeave is developing its own AI model that is not based on other open-source models.
TraceWeave's goal is to build an on-device AI agent that does not require cloud calls. This on-device AI agent can call external tools, interpret the results, and determine the next action. TraceWeave has already built its own demo that runs a multi-step agent using only a smartphone CPU.
In mobility, TraceWeave completed a paid proof of concept (PoC) related to driving data analysis with the autonomous-driving affiliate of a global automaker group. It is currently carrying out performance validation of an in-vehicle AI agent with a global automotive electronics software company. The scope of application is expanding to robots, drones, smart glasses, and wearable devices.
TraceWeave was co-founded in October 2025 by researchers with experience at ETH Zurich, Google, and KAIST. The company raised about KRW 3 billion in a seed funding round led by Mirae Asset Capital. The valuation recognized in this investment was about KRW 30 billion.
Yang Sang-hwan, head of Naver D2SF Center, said that on-device AI is a core technology for implementing complex and advanced AI across a variety of device environments. He also said that on-device AI has advantages in terms of cloud call costs and privacy. He added that TraceWeave is a team that is redesigning AI architecture to implement AI agents in on-device environments.
Yang said that Naver D2SF has identified core technologies for new device environments, such as Nota, a company specializing in AI optimization and lightweighting. He added that Naver D2SF plans to continue investing in startups with related technologies.
Source: TECHWORLD · Kim Seung-gi
Original: https://www.epnc.co.kr/news/articleView.html?idxno=408543
References
This article was produced with the help of an automated content generation algorithm.
Source: TECHWORLD
View originalThis article was summarized and organized by BizCrush based on the original article from TECHWORLD. For exact quotations and full details, please refer to the original article.