Maniformer supplies structured training data to teams building robot-control models and models that predict how physical scenes change. For a hypothetical example, a lab teaching a robot to open doors could browse the Data Marketplace, filter by scene (home, kitchen, office, industrial), data type (real or simulated), robot-arm configuration, and tactile coverage, and purchase an existing dataset for twisting a handle and pushing a door open. If no catalog entry fits, the buyer commissions a custom capture.

The marketplace lists task-specific datasets spanning kitchens (microwave heating, sink cleaning, hanging utensils), bedrooms (door opening), study rooms (laptop opening, bookshelf retrieval), entertainment areas, and industrial and catering environments. Tasks are tagged by teleop mode (single-arm or dual-arm) and whether tactile information is included. The company reports producing more than one million hours of data collected without requiring a robot for every demonstration, although hours alone do not establish how well a customer’s robot will learn a task.

MEgo Gripper is a portable two-finger capture tool that records camera images, motion, gripper opening, and optional touch information while a person performs a task. MEgo View combines head-mounted cameras with wrist cameras so the recordings include both the wider scene and close-up hand-object interactions; the two products can work separately or together.

A person can demonstrate opening a microwave or moving tableware while the devices synchronize their recordings. Maniformer’s cloud tools reconstruct movements, check quality, and add labels, producing structured records that a robotics lab can use to train and test its own model. A human demonstration still needs to be adapted to the robot’s body and controls before it becomes usable robot actions.

Maniformer also maintains a library of 3D object and environment assets for generating synthetic training data in physics simulators, covering household and industrial scenes. These complement real-world captures by enabling low-cost data augmentation.

Maniformer offers ready-made datasets and custom data-collection services to enterprise buyers. Buyers contact the company through business-inquiry forms; the catalog and capture products are offered through direct sales inquiries. Its Hive Data Co-creation Initiative aims to connect suppliers and buyers using shared standards; the initiative does not establish that Maniformer already earns transaction fees from an operating exchange.

The company's main costs are human operators who perform demonstrations, QA annotators who validate trajectory quality, MEgo hardware manufacturing and maintenance, cloud storage, and GPU processing for automated pose reconstruction and labeling. Because each dataset is captured for a specific task and environment, repeat licensing to multiple buyers could improve unit economics if contractual rights permit reuse, but the terms governing exclusivity and resale have not been disclosed. Maniformer's parentage in Agibot provides a natural anchor customer—CEO Yao has stated that Agibot obtains data from Maniformer "through market-based orders"—but this relationship does not guarantee an automatic exclusive data feed, and winning external buyers is important to building a business beyond its parent’s demand.

Maniformer competes with Lightwheel and Config for robot training data budgets, while robot manufacturers can also operate their own collection programs. The main trade-off is whether a buyer wants ready-made recordings, equipment to collect its own demonstrations, or a supplier that also trains and deploys the robot’s control model.

Lightwheel combines physically calibrated simulation assets, first-person human data through EgoSuite, and RoboFinals simulation evaluations. Its simulation and NVIDIA integrations offer buyers a way to create and test training examples before using physical robots, while Maniformer combines a task catalog with MEgo capture hardware and a growing supply network. Maniformer must demonstrate useful data across customers’ different robots rather than competing only on the amount collected.

Config sells converted human demonstrations through Catalog and uses Loop to collect data, train models, and deploy robot policies. It also develops CFG-1, a model for manipulation with two arms, and supports other model providers. Maniformer is positioned as a data supplier rather than a seller of its own robot brain; Config’s broader service can appeal to buyers that want a working task model, while teams developing their own models may prefer to purchase data separately.

Robot companies can gather demonstrations themselves, and public releases such as Agibot World provide training material without a commercial dataset purchase. Internal teams have direct access to their hardware and deployment failures, but must pay for operators, equipment, and quality checks. Maniformer has to offer better coverage or lower collection effort than those alternatives, while respecting the data rights of customers that may compete with Agibot.

Maniformer’s dataset catalog addresses teams training robot models. Its capture devices and China Telecom partnership provide potential routes into equipment procurement and industrial infrastructure budgets, beyond purchases of finished training recordings.

China Telecom's lead investment in the August 2026 round was explicitly tied to a planned joint offering: standardized "cloud-network plus embodied data" solutions targeting humanoid-robot companies, smart-manufacturing OEMs, research institutes, smart-park operators, and smart-home government-enterprise customers. The partnership could combine Maniformer’s collection and data-processing tools with China Telecom’s network, cloud, and enterprise distribution. If the partnership produces a productized bundle—where a factory or smart-park operator purchases data-collection infrastructure alongside telecom connectivity—Maniformer would access industrial-automation budgets that currently flow to system integrators rather than data vendors. This path depends on completing the joint solution, winning actual procurement contracts, and demonstrating that Maniformer's MEgo-captured data improves robot performance in factory settings, rather than treating an announced partnership as a completed deployment.

Maniformer offers MEgo Gripper and MEgo View through its product site. Selling the devices and supporting software more widely to robotics labs and universities could generate a separate revenue stream from teams that want to collect their own data without buying a finished dataset. A competitively priced device and processing package could reach academic labs and smaller robotics startups that need capture equipment but do not want to commission full custom datasets. The incremental buyer here is the research team with its own policy-training pipeline that only needs raw demonstrations and trajectory reconstruction—a different budget line from enterprise dataset purchases. Scaling this channel requires manufacturing MEgo at volume, supporting diverse robot embodiments in the cloud pipeline, and convincing buyers that a capture tool from an Agibot affiliate will serve their needs neutrally.

Data transfer: Maniformer’s recordings must help customers train robots with different bodies, hands, and operating environments. If the data needs extensive adaptation or recollection, a large hour count provides little commercial advantage.

Collection and compliance costs: Human demonstrations require equipment, operators, quality checks, storage, and processing, so expanding supply can consume substantial capital. Collection in factories and homes also requires appropriate permissions and data controls, which can complicate reuse and sales across borders.

Supplier neutrality: Maniformer’s Agibot origin and CEO Maoqing Yao’s dual leadership role may concern robot makers competing with its parent. If buyers are uncomfortable sharing task requirements or proprietary recordings, they could choose another supplier or collect data internally.