Moshi

Further capabilities include an in-app browser preview for opening the host's local development servers, a file browser to explore repository trees at any commit with syntax highlighting, and customizable shortcut panels for tmux, zellij, and agent commands. The app supports hardware keyboards on iPad with dedicated shortcuts like Command-K and Command-O, and offers extensive theming for terminals, chrome, and even app icons. Security features include Face ID for unlocking SSH keys stored in the Keychain, and session recovery that picks up exactly where left off if the app is killed by iOS. It also provides first-class CJK input support, webhook alerts via push notifications for long job completions, and remote clipboard synchronization using OSC 52.

Technically, Moshi works by establishing a direct connection to the user's development machine using the Mosh protocol, which is designed to survive network changes, sleep, and even app termination. This resilient connection ensures that sessions remain alive and can be reattached seamlessly. The app acts as a mobile frontend that communicates with the host, where the AI coding agent is actually running, allowing for real-time interaction. It leverages native iOS capabilities for notifications, Live Activities, Dynamic Island, and Apple Watch integration, enabling approvals and monitoring from anywhere. The architecture is optimized for battery efficiency and low latency, making it suitable for prolonged use.

Benefits for users include the ability to stay connected to development workflows from any location, significantly increasing flexibility and productivity. Measurable outcomes involve reduced downtime, as developers can quickly approve tasks or provide feedback without returning to their computer, and improved monitoring of long-running AI coding sessions. The app's efficient design leads to better battery life compared to traditional remote shells, and its specialized features for AI agents streamline common interactions, saving time and reducing friction. Users gain peace of mind knowing their sessions are persistent and recoverable, even in unstable network conditions.

Concrete use cases include a developer on a train using Moshi with a trackball keyboard to operate Claude Code on their home PC, directing the agent to implement features while commuting. Another example is receiving a push notification on an Apple Watch when a long job finishes, then using the watch to approve the next step without pulling out the phone. A user might take a screenshot of a bug, annotate it directly in Moshi, and paste it into the agent's prompt for debugging, all while away from their desk. Workflows also involve browsing a git repository's history on the phone to review past commits, then switching to a tmux panel to restart a server via a custom shortcut.

Target users are developers who utilize AI coding agents like Claude Code, Codex, OpenCode, Gemini, Cursor, Kimi, or Qwen and want mobile access to their sessions. The app integrates with tools such as tmux, zellij, and herdr, and supports tech stacks involving Mosh, Tailscale for networking, and local speech models. Pricing plans are not detailed in the content, but the app is available on the App Store and Google Play, with a high rating from a community of developers. It appeals to those who value mobility, resilience, and specialized features for AI-assisted coding, whether they are working on personal projects or professional development tasks.

In summary, Moshi unlocks the potential of AI coding agents by providing a robust, mobile-first interface that keeps developers connected to their workflows from anywhere. Its combination of resilient connections, deep terminal integrations, and multimodal input options makes it an essential tool for modern development practices. By addressing the pain points of remote access and battery drain, it enables a seamless transition between desktop and mobile environments, ensuring that AI-assisted coding can continue uninterrupted regardless of location.

Updated 2026-02-28