On June 11, Xiaomi did one thing: it simultaneously launched MiMo Code and open-sourced MiMo SoloEngine.
On the surface, these appear to be two separate products. MiMo Code is a terminal-based programming assistant for developers, while SoloEngine is a low-code Agent platform for everyone.
But if you understand Xiaomi's strategic vision, you'll realize: these are two entry points to the same thing.
I. The Developer's Gateway: MiMo Code
The core breakthrough of MiMo Code lies in its persistent memory system. Traditional AI programming assistants suffer from a "forgetful" problem during long sessions—as the context window fills up, earlier conversations get lost. MiMo Code solves this through a three-layer mechanism: project memory, session checkpoints, and task progress tracking. The main Agent focuses entirely on the work, while a dedicated sub-Agent handles all the recording.
Combined with Compose mode, users simply press Tab, provide a high-level idea, and the system automatically executes the entire pipeline: design → planning → coding → testing → review. It achieves 62% on SWE-Bench Pro, outperforming Claude Code by 5 percentage points.
Developers' problems are solved. They now have an AI programming partner that maintains clarity across hundreds of interaction rounds.
But developers account for less than 1% of the total population.
What about the other 99%?
II. Everyone's Gateway: MiMo SoloEngine
SoloEngine is the first low-code Agentic AI development platform. It tackles the same challenges as MiMo Code—enabling Agentic AI to handle long-horizon tasks, make autonomous decisions, and facilitate multi-agent collaboration—but through an entirely different entry point.
MiMo Code's entry point is the terminal. You need to know how to use the command line, understand programming concepts, and grasp the Harness and Compose modes.
SoloEngine's entry point is a visual canvas in your browser. You drag Agents onto it, connect them with lines, and click run. The backend automatically compiles everything into an executable Agentic AI system.
Two entry points, one underlying capability.
Each Agent in SoloEngine runs the same ReAct loop (think → act → observe → repeat), supports multi-agent collaboration, and employs progressive disclosure to control token consumption. The technical architecture shares the same foundation as MiMo Code—both are genuine Agentic AI, not preset workflows.
| Dimension | MiMo Code | MiMo SoloEngine |
|---|---|---|
| Target Users | Developers | All professionals |
| Entry Point | Terminal command line | Browser-based visual canvas |
| Core Capability | Persistent memory + Compose mode | ReAct + multi-agent collaboration |
| Programming Required | Yes | No |
| Agentic AI | ✓ | ✓ |
| Multi-agent Collaboration | ✓ | ✓ |
| Open-source License | MIT | Apache 2.0 |
III. Why Two Lines Are Necessary
One number explains everything: 87% of enterprises claim to have deployed AI at scale, but only 10% have actually derived value from it.
The root cause of this gap isn't insufficient models or computing power—it's that the ability to build Agents remains locked in developers' hands.
Developers already have Cursor, Claude Code, and MiMo Code—they can work efficiently through Vibe Coding. But a lawyer won't open a terminal, an accountant won't write Python, and an operations manager won't configure a ReAct Agent.
If Xiaomi only built MiMo Code, it would serve only the developer community. If Xiaomi only built SoloEngine, it would abandon the most technically savvy group.
Only by advancing both lines simultaneously can we cover all users.
Developers use MiMo Code for efficient programming via Vibe Coding in the terminal. Professionals from all other industries use SoloEngine to build their own Agent teams through visual operations in the browser.
Both lines share the same model foundation—MiMo. Both lines share the same technical philosophy—genuine Agentic AI. Both lines share the same goal—enabling AI to complete tasks autonomously rather than following preset paths.
IV. SoloEngine's Unique Value: Vertical Deep Dive
MiMo Code is a general-purpose programming assistant—it can write any code and handle any project.
SoloEngine takes a different approach—vertical deep dive.
Manus and the viral "Lobster" earlier this year attempted the general-purpose Agent route—capable of everything but excelling at nothing. General-purpose Agents struggle to meet commercial requirements.
SoloEngine enables experts in every industry to create precise, reusable Agentic AI tools tailored to their specific tasks. A lawyer's Agent handles only legal work, an accountant's Agent focuses solely on financial analysis, and an operations manager's Agent deals exclusively with campaign planning.
Vertical means precision, and precision means usability.
A "Contract Review Agent Team" built by a lawyer using SoloEngine isn't a generic document analysis tool—it's specifically optimized for contract review scenarios, with understanding of legal terminology, familiarity with common risk points, and knowledge of how to cite relevant laws. This vertical depth is unattainable for general-purpose Agents.
SoloEngine's progressive disclosure mechanism makes this vertical deep dive cost-effective. Tools, Skills, and MCP protocols load on demand, reducing token consumption by over 85%. A vertical Agent team operates at a fraction of the cost of a general-purpose Agent.
V. Xiaomi's Agent Ecosystem Closed Loop
From the MiMo foundation model to MiMo Code to MiMo SoloEngine, Xiaomi is constructing a complete Agent ecosystem closed loop:
Brain Layer—MiMo Foundation Model, trillion-parameter scale, global open-source leader. After a 99% price reduction, API costs are extremely competitive.
Programming Agent Layer—MiMo Code, targeting developers, with persistent memory and Compose mode, achieving 62% on SWE-Bench Pro.
General Agent Layer—MiMo SoloEngine, targeting all industries, featuring low-code, visual canvas, and multi-agent collaboration.
Application Layer—miclaw mobile Agent, Agent ecosystem platform, 1 billion IoT devices.
Distribution Layer—SoloEngine's one-click packaging feature enables built Agent teams to be packaged into standalone products for self-deployment or distribution and sales.
From building models to building platforms to building services, Xiaomi's dual-Agent strategy is transforming 60 billion yuan in investment into tangible business value.
MiMo Code is the Agent for developers. MiMo SoloEngine is the Agent for everyone. Two lines, one ecosystem, covering all industries.
Top comments (0)