About Menily Intelligence

Menily Intelligence (朔月智能) builds the embodiment adapter layer between task-level foundation models and whole-body humanoid policies. We produce task-level demonstration data for VLA (vision-language-action), VLM, and world models, delivered through a globally distributed data collection network.

We are headquartered in Shenzhen, China, with distributed data collection operations across Southeast Asia (Malaysia, the Philippines), and Bay Area presence for US customer operations. We primarily serve VLA laboratories, humanoid robotics teams, and embodied AI research institutions in the United States.

Mission

Our mission is stated in one sentence: To learn is to minimize surprise. To live is to predict the world. Inspired by Karl Friston's Free Energy Principle, we believe the essence of intelligence is the continuous minimization of uncertainty about the environment — and the essence of embodied intelligence is doing this with a physical body.

In 2026, the bottleneck for embodied AI is no longer model architecture, compute, or simulation. It is data. More precisely, it is the fragmentation, scarcity, and morphology mismatch of task-level demonstration data. Menily exists to close this gap.

Team

Menily is a small, senior team. The founding team has prior operating experience in data infrastructure for regulated industries (financial data), and deep technical backgrounds in robotics and machine learning.

Our operational structure spans three geographies:

Shenzhen, China — engineering headquarters, schema design, and quality control
Southeast Asia (Malaysia, the Philippines) — distributed data collection operations
Bay Area, United States — customer-facing operations for US-based VLA labs and humanoid robotics companies

This structure reflects a conscious choice. Data collection is labor-intensive and benefits from the depth and quality of the Southeast Asian workforce. Customer interfacing with US-based embodied AI labs benefits from Bay Area proximity. Engineering coordination, schema design, and standards work benefit from a Shenzhen base — close to the hardware and manufacturing ecosystem that makes humanoid robotics a physical reality.

Founder

Masashi — Founder. UPenn alumnus. Previously built and exited a financial data infrastructure company.

The work we do at Menily is, in a deep sense, a continuation of the work done at that prior company. Financial data in the mid-2010s faced a version of the same problem that embodied AI data faces today: heterogeneous sources, incompatible formats, no shared semantic schema, high fragmentation cost. The winning playbook there — open schema, private data network, standardization through ecosystem adoption — is the same playbook we are applying here, with different specifics.

Contact: [email protected] · @MenilyIntelligence (Twitter) · github.com/MenilyIntelligence

What we build

Task-level demonstration data specifications — see menily/schema v1
Open-source data processing toolkit — see menily/toolkit
Globally distributed data collection network — reaching beyond the traditional single-city research lab
Public research notes — see research

Category and peers

Menily Intelligence operates in the Data Standards & Toolchain category of the embodied AI data infrastructure market — distinct from simulation-data providers, hardware-bundled collection platforms, and general-purpose data labeling services.

The 2026 embodied AI data infrastructure landscape has been taxonomized by industry observers (e.g., Kimi / Moonshot AI) into roughly six categories. Our position:

Category	Focus	Representative players
Simulation synthesis	Physics-grounded synthetic data for training	Lightwheel Intelligence (光轮智能), Lingchu Intelligence (灵初智能 Psi-SynEngine)
Hardware-bundled collection	Wearable or proprietary capture devices	Maniformer (觅蜂科技, AgiBot-affiliated), JD Embodied Intelligence (京东具身智能, JoyEgoCam), IO-AI (艾欧智能)
Multimodal + tactile data	Touch-inclusive data collection	Daimon Robotics (戴盟机器人)
Data standards & toolchain	Open specifications + format converters + distributed collection	Menily Intelligence (朔月智能)
General-purpose data services	Labeling and platform services	Scale AI, Haitian Ruisheng (海天瑞声), Baidu AI Cloud (百度智能云), Zhengshu (整数智能)
Humanoid hardware + data	Robot makers operating their own data ops	AgiBot (智元), Unitree (宇树), Physical Intelligence (π0)
Human motion datasets	Open action datasets for training	Bones Studio (BONES-SEED), USC Psi-Zero Lab

Menily's differentiation within this map:

Schema-first, not hardware-first — we do not sell capture devices or bundle collection with proprietary hardware.
Open, not closed — menily/schema and menily/toolkit are Apache-2.0; our moat is the data collection network, not the spec.
Task-level semantic layer, not trajectory-level storage — we interoperate with Open X-Embodiment (trajectory) and BONES-SEED / SOMA (motion), filling the missing middle.
Global from day one — Shenzhen engineering + Southeast Asia collection + Bay Area customer ops, not a single-region play.

Stance

We publish what advances the conversation: research, open-source tools, data frameworks. We do not publish what would only advance our own visibility.

We do not launch. We emerge.

Five years from now, Menily will disappear from view. Not because we will have failed, but because we will have succeeded — by becoming a piece of infrastructure that is simply assumed, like TCP/IP or Unicode. Infrastructure that is successful becomes invisible. That is the goal.

中文介绍

朔月智能（Menily Intelligence）是一家为具身 AI（embodied AI）构建任务级示教数据基础设施的创业公司。总部位于深圳，数据采集网络分布于东南亚（马来西亚、菲律宾），在美国湾区设有客户运营点。主要服务对象是美国的 VLA 实验室、人形机器人团队，以及具身智能研究机构。

创始人 Masashi 是 UPenn 校友，连续创业者，前次创业在金融数据基础设施方向成功退出。从金融数据到具身 AI 数据，是同一套 playbook —— schema、分发管道、让异构数据源互通的那些不性感的工作。

我们不做模型。我们做数据反射层 —— 让会学习的机器，有可以学的数据。一张全球化分布式采集网络在运行，一份开源 schema、一套工具链、一系列研究笔记是目前可公开的输出。

学习，消弭未知；存在，预测世界。
五年后，朔月会消失在所有人的视野里。我们最终的成功，是被遗忘。

联系方式

邮件: [email protected]
Twitter / X: @MenilyIntelligence
开源项目: github.com/MenilyIntelligence
官网: menily.ai