What Xiaomi Miloco 2.0 Can Do: A Deep Dive into the New AI Home Assistant

Today Xiaomi officially released Miloco 2.0, an open‑source AI solution for whole‑home intelligence built around the MiMo large model. Miloco 2.0 is packaged as an Agent that connects to OpenClaw and can be installed on macOS, Linux or Windows systems, with future support for other Claw‑type products.

Miloco 2.0 treats cameras as eyes, microphones as ears, and Mi Home devices as hands, while the Miloco core provides perception, memory and decision‑making so that all devices can cooperate.

Two core evolutions

Long‑term task handling : Unlike the one‑command‑one‑action model of traditional smart homes, Miloco 2.0 can continuously monitor, record status, evaluate conditions and trigger reminders or interventions. Example tasks include:

Alert when a child’s posture becomes slouched.

Automatically turn off the TV after one hour of continuous watching.

Notify if an elderly family member has not returned to their room for an extended period.

Common‑sense memory and proactive service : Miloco 2.0 not only understands what is happening, it also remembers habits and proactively offers services, such as:

Detecting a child’s cry, a possible fall, or a stove left on and assigning a risk level before notifying the user.

Remembering that the user returns home at 8 pm and automatically turning on warm lighting and playing favorite music.

These habits accumulate into a family archive that can generate personalized automation tasks with a single click.

From Miloco 1.0 to 2.0

Miloco 1.0, released in November of the previous year, relied on rule‑driven interactions: users had to pre‑define each rule, which cannot cover the endless variability of real life. Miloco 2.0 upgrades the concept in four ways (multimodal perception, proactive service, family memory, continuous tasks) and introduces four core capabilities:

Rule evolution : From instant triggers to full‑cycle tracking with different responses at start, during, and end of a task.

Scheduling : From single execution to flexible timing (delayed, periodic, timed) that drives automatic reminders and summaries.

Status persistence : From stateless operation to durable archives that support ongoing task tracking.

These three abilities act like a to‑do list, calendar and work log for the AI household manager.

Deployment and cost

To use Miloco 2.0 you need:

A computer capable of running OpenClaw (Mac, Windows or Linux).

A Xiaomi account linked to your Mi Home devices.

At least one Mi Home camera.

An API key for a multimodal large model.

“Help me install the Miloco plugin: https://github.com/XiaoMi/xiaomi-miloco”

OpenClaw then prepares the environment, binds the account and the model, and completes the installation with almost no manual steps.

Miloco 2.0 itself is completely free and open‑source for non‑commercial use. The only costs are the hardware you already own (camera, Mi Home devices) and the API usage fees for the large model. Xiaomi offers a token‑gift promotion for the MiMo model during the first week of release.

Privacy guarantees

The core privacy principle is “process‑then‑discard”: raw video streams never leave the device, only semantic results are sent to OpenClaw. All data are stored locally, cleared after 30 days, and users can configure black‑lists for cameras and white‑lists for households.

Current limitations

Miloco 2.0 still struggles to achieve perfect accuracy in complex, ever‑changing home environments, and the team plans ongoing optimization.

Nevertheless, Miloco 2.0 demonstrates a new lifestyle where smart home systems move beyond executing commands to truly understanding and assisting daily life.