NAVER D2SF announced on October 20, 2026, that it has made a new investment in TraceWeave, an on-device AI startup led by CEO Jaeyoung Park. TraceWeave is the team behind Phase-TITAN, a post-Transformer architecture that understands long contexts and carries out multi-step tasks while keeping memory usage constant. NAVER D2SF recognized the company's core technology for enabling AI agents on resource-constrained devices.
The On-Device Memory Bottleneck
As AI expands to robots, vehicles, smart glasses, and drones, its applications are advancing beyond simple chatbots toward complex reasoning and decision-making. In Transformer-based models, the key-value cache grows with context length, requiring more memory as tasks become longer and more involved. On-device constraints in memory, battery life, and heat have made it difficult to implement AI agents that understand long contexts and perform complex tasks.
Phase-TITAN and Constant Memory Usage
Phase-TITAN is a new AI architecture built to eliminate this memory bottleneck by replacing the conventional Transformer design. TraceWeave develops its own AI models rather than relying on existing open-source models, and memory usage remains constant as context grows. According to TraceWeave, at a 1M-token context without quantization, a comparable Transformer's KV-Cache holds tens of gigabytes or more, while its 1B Phase-TITAN model holds 0.04 GB.
Benchmarks and Multi-Step Agents
This enables agents to call tools, interpret results, and decide on their next actions entirely on the device, without cloud calls. TraceWeave's internal benchmarks on RULER and BFCL indicate that Phase-TITAN maintains performance on par with Transformer models of the same size and larger. The company has demonstrated a multi-step agent running on a smartphone using only its CPU, without cloud calls.
Early Commercial Validation
In mobility, TraceWeave completed a paid pilot with the autonomous driving affiliate of a global automotive group, identifying driving-data frames that improve the partner's autonomous driving model without human annotation. It is also validating in-vehicle agent performance with a global automotive electronics software company and exploring opportunities in robots, drones, smart glasses, and wearables. TraceWeave was a Top 20 finalist in NVIDIA's 2026 Inception Grand Challenge.
Funding and Team Background
TraceWeave was co-founded in October 2025 by researchers with backgrounds at ETH Zurich, Google, and KAIST. The team combines engineering expertise grounded in computational neuroscience and physics with business development capabilities, designing its own architecture and validating it in device environments that reflect real-world customer needs. Based at NAVER D2SF in Gangnam, TraceWeave raised approximately KRW 3 billion in this Seed round led by Mirae Asset Capital at a pre-money valuation of approximately KRW 30 billion.
Executive Perspectives
Jaeyoung Park, co-founder and CEO of TraceWeave, said that removing the KV cache was never the hard part and that staying capable without it was the real challenge. He added that Phase-TITAN already delivers Transformer-class capability without the Transformer's memory tax, and the next model goes well beyond its class with memory still fixed. Sanghwan Yang, Head of NAVER D2SF, described on-device AI as a core technology for complex, advanced AI across diverse device environments, offering advantages in reducing cloud call costs and protecting privacy.
TraceWeave's constant-memory approach addresses one of the central technical barriers to deploying sophisticated AI agents on resource-constrained hardware. Its early traction in smartphones, mobility, and automotive validation suggests practical interest across industries that need local reasoning without cloud dependency. With new Seed funding and a research team focused on next-generation architecture, the company is positioned to expand its commercial footprint and advance on-device AI.