MiniMax H3-Image
MiniMax's image generation and editing model supports up to 16 reference images and high-resolution output, offers an API and an open-weight approach, and is suitable for exploring products, characters, and design concepts.
Pinky: Let AI operate the phone, while retaining human verification and the ability to take over
Pinky is an open-source mobile automation tool developed by DIYgod. Through MCP, it allows agents such as Claude Code, Codex, and Cursor to connect to real Android phones or iPhones to perform actions like viewing the screen, tapping, and typing. Users can monitor the process on their computer, verify key steps, and take over at any time, making it ideal for those who want to incorporate mobile tasks into their agent workflows.Official Project Description
Image source:DIYgod/Pinky Official Project README, Not tested by this site. The project uses Apache 2.0 License.
What can I do?
In addition to connecting to existing agents, Pinky can also perform tasks using models selected by the user. Official documentation is available for both desktop and mobile platforms, and new features—such as task scheduling—are being added on an ongoing basis. The software is free and open-source; model invocation fees depend on the selected service. Task execution is subject to device specifications, system versions, the application interface, and model capabilities, and should not be interpreted as a guarantee that all mobile operations can be reliably completed.Chinese Update Log
The official repository, documentation, and installation packages are publicly available, with versions for macOS, Windows, and Linux. The release page for v2026.1115.1 is dated November 14 at 16:45 UTC, which is November 15 at 00:45 Beijing Time; this is a maintenance update, not the project’s initial release.Official Version Page
How to Get Started
- Select the installation package that matches your computer's operating system from the official release page, and be sure to read the connection requirements for your phone first.
- Follow the on-screen prompts to pair your device, then connect to an MCP-compatible agent, or configure the model of your choice.
- Start by testing the process with low-risk tasks, such as opening an app or looking up known information, observing each step as you go, and then gradually expanding the scope of use.Getting Started and Compatibility Information
Who Is It For?
It is better suited for developers, mobile app testers, and advanced users who are willing to configure devices. For example, you can check a sequence of actions on a test phone or have an agent help locate information within an app, with a human then verifying the results. When testing multiple device models, you should also evaluate device management and concurrency requirements separately.
Precautions Before Use
The project is still in the pre-alpha stage, and its features and compatibility are subject to rapid change. The setup process varies depending on the iOS version, and some configurations require a Mac and Xcode. We recommend testing the app first on a spare device and with non-sensitive tasks. For actions involving payments, sending messages, or deleting content, ensure manual confirmation is required, and do not treat unattended operation as the default behavior.
How to Choose Among Similar Options
| Tools and Positioning | Features and Use Cases | Usage Restrictions |
|---|---|---|
| Pinky: A Local Mobile Automation App | MCP Access, Viewing, and Takeover—Perfect for Your Device | Pre-alpha; pairing required; compatibility depends on the system version |
| mobilerun framework: Open-source framework | Write and integrate your own mobile automation workflows | Python Environment and Device Configuration Required |
| mobilerun Cloud: Cloud Device Services | APIs, cloud phones, and device management—ideal for multi-device tasks | A cloud account and the corresponding device environment are required. |
When making your selection, first decide whether to use your own phone or a cloud-based device, and then check system compatibility and model costs.
data statistics
Relevant Navigation
Shortest
An end-to-end testing framework based on natural language processing and AI technologies which streamlines the testing process, increases testing efficiency, and lowers the testing threshold.
AlphaDrive
Combining visual language modeling and reinforcement learning, the autopilot technology framework is equipped with powerful planning inference and multimodal planning capabilities to deal with complex and rare traffic scenarios.
BabelDOC
Open source AI translation tool, supporting bilingual control, multi-engine translation, format preservation and batch processing, helping researchers read foreign literature efficiently.
Toonflow
An open-source AI short-form drama production platform—a fully automated pipeline that takes a novel and turns it into a finished video. It features skeleton binding to prevent face distortion, produces an episode in 8 minutes, and keeps costs down to just over ten yuan.
DeepSeek-V4
The V4 model series, developed through in-depth research, includes V4.1-Flash—which supports text-and-image understanding, long context, and tool invocation—as well as V4 Pro, which continues to provide services.
PaddleOCR-VL
Baidu's lightweight multimodal document parsing model, with 0.9B parameters, achieves accurate recognition and structured output of complex documents in 109 languages, with world-leading performance.
Mika
Mika converts natural language tasks into mobile and web operations, supporting virtual screens in the background, task playback, and custom model selection; it is suitable for low-risk automated exploration, but users should carefully review permissions, data flow, and verification mechanisms before use.
No comments...
简体中文 
English