Control AI Coding Agents from Your iPhone: A Guide to the Agmente Project

Published on ClawList.io | Category: AI Automation | Mobile Developer Tools

---

If you've ever wished you could manage your AI coding sessions from your couch, commute, or anywhere away from your desk, the Agmente open-source project is about to change the way you think about mobile developer workflows. Built by Lü Peng (@lvpeng), an Engineering Manager on the VS Code team at Microsoft, Agmente bridges the gap between powerful AI coding agents and the convenience of your iOS device.

As AI-powered tools like Claude Code, Gemini CLI, and Qwen become central to modern software development, the ability to supervise, direct, and interact with these agents on the go represents a genuine leap forward in developer productivity. Thanks to projects like Agmente — and the broader momentum from OpenClaw — we're entering a new era where your iPhone becomes a legitimate command center for autonomous coding workflows.

---

What Is Agmente and Why Does It Matter?

Agmente is an open-source project that allows developers to interact with AI coding agents — commonly called Coding Agents — directly from an iOS device. Instead of being tethered to a laptop or desktop machine to supervise an agent's activity, you can:

• Start and direct conversations with agents like Claude Code, Gemini CLI, and Qwen
• Monitor real-time tool calls — see exactly what commands the agent is executing
• Review execution results as they happen, in a clean mobile interface
• Stay in control of autonomous AI workflows without sitting at a workstation

The project was inspired in part by the OpenClaw ecosystem, which has demonstrated many compelling scenarios where developers issue high-level instructions from their phones and let agents handle the heavy lifting — writing code, running tests, querying databases, and more.

What makes Agmente particularly valuable is its real-time visibility into agent behavior. One of the biggest concerns developers have about autonomous coding agents is the "black box" problem — you hand off a task and have no idea what the agent is doing until it's done. Agmente solves this by surfacing tool call logs and execution outputs directly in the mobile UI, giving you a transparent window into the agent's decision-making process.

---

How Agmente Works: Architecture and Key Features

At its core, Agmente acts as a mobile interface layer that connects to AI coding agents running on a backend (typically your development machine or a remote server). Here's a breakdown of how the system operates:

Supported AI Agents

Agmente currently supports interaction with several leading AI coding agents:

• Claude Code — Anthropic's agentic CLI tool for software engineering tasks
• Gemini CLI — Google's command-line AI agent powered by Gemini models
• Qwen — Alibaba's powerful open-source language model ecosystem

This multi-agent support means you're not locked into a single provider. You can switch between agents depending on your task — for example, using Claude Code for complex refactoring and Gemini CLI for rapid prototyping.

Real-Time Tool Call Visibility

One of Agmente's standout features is its live tool execution feed. When an agent runs a tool — such as reading a file, executing a shell command, or making an API call — Agmente displays this activity in real time. A typical interaction might look like this:

[Agent: Claude Code]
> Task: Refactor the authentication module to use JWT tokens

[Tool Call] read_file("src/auth/login.js")
[Result] File read successfully — 142 lines

[Tool Call] write_file("src/auth/login.js", <updated_content>)
[Result] File written successfully

[Tool Call] run_command("npm test -- --testPathPattern=auth")
[Result] ✅ 12 tests passed, 0 failed

[Agent Response] Authentication module has been refactored. JWT implementation
is in place and all tests pass. Consider updating your .env file with
JWT_SECRET variable.

This level of transparency is invaluable. You can catch mistakes early, intervene if an agent is headed in the wrong direction, or simply stay informed while you're away from your desk.

iOS-Native Mobile Interface

The iOS interface is designed with developer ergonomics in mind:

• Conversation view for natural language interaction with agents
• Tool call log panel for real-time execution monitoring
• Session management to switch between multiple agent sessions
• Notification support so you're alerted when an agent completes a task or needs input

---

Practical Use Cases: Commanding Agents from Your Phone

The Agmente project unlocks a set of genuinely useful real-world scenarios that were previously impractical. Here are a few examples:

1. Kicking Off a Coding Task During Your Commute

Imagine you're on your morning commute and remember a bug that needs fixing before the team standup. With Agmente, you can:

You: "Fix the null pointer exception in the user profile endpoint.
      The error occurs when the avatar_url field is missing."

Agent: Starting investigation...
[Tool Call] search_codebase("avatar_url")
[Tool Call] read_file("src/controllers/userController.js")
[Tool Call] write_file(...)
[Tool Call] run_tests(...)

Agent: Bug fixed. The null check has been added and tests confirm
       the endpoint now handles missing avatar_url gracefully.

By the time you arrive at the office, the fix is ready for review.

2. Monitoring Long-Running Agent Tasks

For tasks that take 10–30 minutes — like scaffolding a new microservice, migrating a database schema, or generating documentation — Agmente lets you check in periodically on your phone rather than watching a terminal screen the entire time. You get progress updates, tool call logs, and a notification when the job is done.

3. Collaborative Pair Programming with AI

You can have an ongoing dialogue with the agent, refining its approach step by step:

You: "The API structure looks good, but let's switch from REST to GraphQL
      for the data-fetching layer."

Agent: Understood. I'll refactor the relevant endpoints...

This kind of iterative, conversational control is exactly what makes modern AI coding agents so powerful — and Agmente brings it to your fingertips, literally.

---

Getting Started with Agmente

To try Agmente yourself, start by checking out the open-source repository. Since it's an evolving project, the best way to follow updates is to:

• ⭐ Star the GitHub repository to stay updated on new releases
• Follow Lü Peng on X/Twitter for development updates and feature announcements
• Join the OpenClaw community to explore broader mobile AI agent workflows

While specific installation steps will vary as the project evolves, the general setup involves:

# Clone the repository
git clone https://github.com/lvpeng/agmente

# Follow the setup guide in the README to connect your iOS device
# to your preferred AI coding agent backend

Make sure your development environment has your chosen agent (Claude Code, Gemini CLI, or Qwen) properly configured before connecting Agmente.

---

Conclusion: The Future of Mobile AI Development

The Agmente project represents something more significant than just a convenient mobile interface — it's a glimpse into the future of how developers will interact with AI agents. As these agents become more capable and autonomous, the need to monitor, guide, and collaborate with them from anywhere becomes increasingly important.

Lü Peng's work — informed by his deep experience on the VS Code team — brings a thoughtful, developer-first design philosophy to the mobile AI agent space. Combined with the ecosystem momentum from OpenClaw, projects like Agmente suggest that the "AI pair programmer" of tomorrow won't require you to be at a desk at all.

Whether you're a full-stack developer looking to manage agents during off-hours, an AI engineer building automation pipelines, or simply an enthusiast exploring the frontier of AI tooling, Agmente is worth adding to your toolkit.

Your iPhone is no longer just a communication device — with Agmente, it's a coding agent command center.

---

Original source: @dotey on X/Twitter Project by Lü Peng, Engineering Manager at the VS Code Team Tags: #ClaudeCode #GeminiCLI #AIAgents #MobileDev #OpenClaw #Agmente #AIAutomation