TraceWeave raises $2.27M seed to run AI agents on-device
What's the deal? On-device AI startup TraceWeaveDealroom has a profile for this one. Try Dealroom → has raised approximately US$2.52M (roughly $2.27 million) in a seed round led by Mirae Asset CapitalDealroom has a profile for this one. Try Dealroom →, with participation from NAVER D2SF. The deal, announced in October 2026, valued the company at a pre-money US$25.2M.
What's the endgame? TraceWeave builds its own post-Transformer architecture, Phase-TITAN, designed to let AI agents run on memory-constrained devices. The company develops its own models rather than optimising existing open-source ones, keeping memory use constant as an agent's context grows.
The technical pitch targets a known bottleneck. In Transformer models, the key-value cache expands with context length, straining device memory, battery, and heat. TraceWeave says that at a 1M-token context, a comparable Transformer's cache holds tens of gigabytes, while its 1B Phase-TITAN model holds 0.04GB — enough to let agents call tools and decide next steps entirely on-device, without cloud calls.
Why now? As AI expands to robots, vehicles, smart glasses, and drones, applications are moving beyond chatbots toward complex reasoning. Local processing can cut latency, reduce cloud costs, and keep sensitive data on the device.
TraceWeave, co-founded in October 2025 by researchers with backgrounds at ETH Zurich, Google, and KAIST, has early traction. It completed a paid pilot with the autonomous driving affiliate of a global automotive group, is validating in-vehicle performance with a global automotive electronics software firm, and was a Top 20 finalist in NVIDIA's 2026 Inception Grand Challenge. It plans to use the funding to expand its research and engineering team and grow its business.
"On-device AI is a core technology for implementing complex, advanced AI across diverse device environments beyond PCs, and it also offers advantages in reducing cloud call costs and protecting privacy," said Sanghwan Yang, head of NAVER D2SF.
What could go wrong? This is an early-stage endorsement, not proof of commercial adoption or technical advantage at scale. TraceWeave's performance claims rest on its own internal benchmarks, and it is betting on replacing the Transformer rather than optimising it — a far harder path.
The signal: The round sits well above typical seed sizes, placing it in the upper tier for the stage. More telling is where the bet lands: at the model-architecture layer of the on-device stack, rather than on compressing existing models as peers like Korea's Nota AI do. With edge-AI activity rising — Edgify recently raised $9 million to expand local processing — investors are starting to fund the foundations, not just the optimisations.
Read more: prnewswire.com
Image credit: Lenharth Systems
