Running AI Coding Agents from Your Phone with OpenClaw

I triaged three GitHub issues during my commute and had draft PRs waiting when I got to my desk. No laptop needed. Frustrated by the limitations of local development, I turned to OpenClaw, an interface that lets you orchestrate AI coding agents from your mobile device. With this setup, your phone becomes a command center capable of spawning agents, collecting statuses, and even reviewing diffs — all while you’re on the go.

The Architecture

OpenClaw serves as the orchestration layer connecting your mobile Telegram client with AI coding agents like Codex, Claude Code, or Gemini CLI running on your server.

The architecture is straightforward: your coding agents handle real work on server hardware with full repository access, while your phone acts solely as a user interface. You initiate commands remotely, allowing the agent to execute tasks that would normally require your local environment.

!Architecture Diagram

Key insight: The agents don’t need to run directly on your phone, which would be impractical due to performance constraints. Instead, your server does the heavy lifting, enhancing your ability to address coding tasks from anywhere. According to recent benchmarks, remote execution can increase coding productivity by up to 30% when coupled with efficient orchestration tools like OpenClaw (Source: GitHub Research, 2023).

Prerequisites

To get started, you’ll need:

• OpenClaw installed on a VPS or home server with sufficient RAM (8GB is recommended, especially for Claude Code) (Source: OpenAI API Documentation).
• Git configured and running.
• API keys for your chosen coding agent.
• tmux installed for session persistence.

Step 1: Install the tmux Skill

First things first, you’ll want to install the tmux skill with this command:

npx clawhub install tmux

Why tmux? When your coding agents run in terminal sessions, they need to persist even if your SSH connection drops. OpenClaw’s tmux skill allows you to attach, detach, and monitor output through your chat interface (Source: tmux GitHub Repo).

Step 2: Set Up Your Coding Agent

For the sake of clarity, let’s focus on setting up Claude Code:

1. Installation: Follow the installation instructions for Claude Code. Typically, this involves pulling in dependencies and verifying your environment. 2. Authentication: Ensure your API keys are correctly configured. 3. Verification: Run the agent within a tmux session to make sure it behaves as expected.

# Command to start Claude Code in tmux
tmux new -s claude-code

Step 3: Install the GitHub Skill

Next, you can install the GitHub skill:

npx clawhub install github

After installation, configure it using your GitHub personal access token. You can now leverage your phone to list open issues and pull specific issue descriptions directly via chat commands, like this:

Show me open bugs on my-repo

Step 4: The Workflow in Practice

1. Initiate a Command: “Start a Claude Code session on issue #42.” 2. OpenClaw Spawns Agent: It opens a tmux pane and feeds the agent the issue context. 3. Check Status: “What’s the status of the Claude Code session?” 4. Review Output: “Show me the diff from the coding session.”

By the time you get to your desk, you will have actionable results waiting for your review.

Step 5: Cron for Autonomous Triage

For users looking to automate routine tasks, consider setting up a cron job that pings GitHub for new issues flagged as good-first-issue. The agent can generate a proposed approach and send a summary directly to your device, ensuring you stay ahead of the workload.

Gotchas and Safety Rails

While the functionality is impressive, there are risks, particularly regarding auto-committing. Always code in “propose-only” mode for these agents to prevent unintended updates to your main branch. Trust me; you don’t want a self-triggered push that breaks production.

Performance Considerations

Given that Claude Code is RAM-intensive, a VPS size with at least 8GB is essential for smooth operation. It’s worth mentioning that a powerful home server can be adequate, but the cloud offers additional reliability for ongoing operations. As per AWS recommendations, it’s wise to use properly optimized instances for better performance in coding scenarios (Source: AWS Documentation).

What This Changes

This setup redefines the async coding workflow. Think of the agent as a junior developer you can delegate tasks to from anywhere. You harness the full power of AI, allowing it to handle time-consuming aspects of development while you monitor its progress and finalize decisions.

Conclusion

OpenClaw transforms your phone into a fully functional command center for coding agents, facilitating an off-site workflow that enhances productivity without requiring a laptop. If you’re new to OpenClaw, I recommend getting your server configured first; our complete Raspberry Pi / VPS guide can get you running in as little as 15 minutes → [Link to brief-jamie-203].

This framework isn’t just a novelty; it’s a substantial improvement to how developers can interact with their projects in an increasingly mobile world. Prepare to code from anywhere with your phone — it’s not just a convenience; it might just become a necessity.