The Venture Codex

Led · Equity · Aug 2026

KNOWIN, founded about a year ago, is developing home robotics hardware and software including the NOWIN-X1, a foldable, wheeled dual-arm robot. The company pairs that hardware with its GLOW world model to enable complex household task execution. After securing roughly $74 million in funding, KNOWIN is focused on scaling operations and progressing toward mass production. The financing is presented as accelerating the startup’s transition from development to commercialization. The article does not provide revenue, user, or valuation metrics.

Participated · Equity · Aug 2026

Founded in 2024, Sand.ai develops foundational models for video generation and audio-visual synchronization, including its Magi-1 video generation model and Gaga-1 audio-visual sync model. The company also builds digital humans and AI video agents and has expanded into application-specific offerings such as a music-generation agent called VidMuse. VidMuse reached about $10 million in annual recurring revenue roughly three months after its early-2026 release. Sand.ai says its Magi-1 ranks highly on DeepMind's Physics-IQ benchmark, and it plans to release a next-generation video-generation model in the July–September quarter that will use a Mixture-of-Experts architecture and aim to be a large open-source model with high inference efficiency. The company faces competition from players such as Kuaishou's Kling, ByteDance's Seedance 2.0 and Runway.

Participated · Series B · Aug 2026

Founded in 2023, Aureka builds large biological foundation models (AuraIDE and its open-source counterpart OpenDDE) and integrates them with Lab-in-the-Loop — a closed-loop system combining AI agents, digital biology and high-throughput experimental platforms. Its foundation models are trained on proprietary protein co-evolution data and evaluated as among the leaders in protein structure prediction and de novo molecular design. The company reports generating tens of millions in revenue over the past two years and has established strategic partnerships with multiple global pharmaceutical companies. Aureka’s platform couples model-driven molecular generation with in-house single-cell functional screening and automated wet-lab validation, feeding experimental results back into model post-training to iterate rapidly. With the Series B proceeds, Aureka plans to scale model training, upgrade its experimental feedback engine, accelerate validation and translate model capabilities into antibody drug assets.

Participated · Series A · Jul 2026

RoboParty produces a fully open-source two-legged humanoid robot prototype and an accompanying open development platform that enables self-assembly, action replication, and algorithm debugging. Since open-sourcing, its GitHub project has received about 1.7K stars and its global developer community has exceeded 5,000 contributors. The team has shipped the first several hundred devices to early adopter universities, labs and developers, leading to high purchase and usage rates and repeat orders, with many users preordering the next product, RP1. RoboParty has earned industry recognition including third place globally at the ICRA Reasoning to Action competition and multiple open-source awards in embodied intelligence. The company is building a young, full-stack technical team recruiting from top universities worldwide and plans to continue iterating on ontology structure, motion control, documentation, development tools and hardware. It aims to scale mass replication and industrial deployment of humanoid robots by leveraging open source, youthful talent, and real user scenarios.

Led · Seed · Jul 2026

Lingjing Zhiyuan develops a full-stack, hardware-software integrated computing infrastructure and native operating system (MS-OS) for embodied intelligence, aiming to unify models, chips, and systems for various robot forms. The company has adopted an in-house, capital-intensive R&D strategy that includes a heterogeneous hardware computing solution (“Dvořák”), a brain-inspired collaborative model system, and a domestically developed MS-OS compatible with multiple robot platforms. Founded in July 2025 and incubated by the Shanghai Artificial Intelligence Research Institute, its team includes founders and core members with backgrounds at Huawei, ZTE, and Kunlunxin, and it benefits from research and state-capital relationships. In its first year the company has connected with over 300 embodied system manufacturers and reports more than 200 customers that have completed mass deployment or placed small-batch orders, with some customers ordering over 1,000 units. Technically, Lingjing Zhiyuan says its FPGA prototype has completed full-process verification and it plans to begin dedicated chip design in mid-2026 with tape-out planned for 2027. The company positions itself to address real-time perception-decision-execution bottlenecks in robotics and intends to use continued financing to support R&D, mass production ramp-up, and ecosystem and standards development.