How to Monitor AI Builds from Your Phone

You kicked off three AI agents before leaving for lunch. One is scaffolding a new API, another is writing integration tests, and a third is refactoring a legacy module. Now you are walking to a restaurant and want to know if everything is still running smoothly.

This is the exact scenario Remocode was built for. Here is how to monitor AI builds from your phone, step by step.

The Setup: Remocode + Telegram

Before you leave your desk, make sure Remocode is running with your workspace loaded and Telegram connected. Each AI agent should be running in its own pane with a descriptive name like api, tests, or refactor.

Once Telegram is linked, your phone becomes a live window into every terminal session on your machine.

Quick Status Check with Peek

The fastest way to check on an agent is the peek command. Open Telegram and type:

peek api

Remocode sends back the last visible lines from your api pane. You can see exactly what the agent last output without scrolling through a full log. It takes two seconds and costs nothing.

Checking Multiple Panes

Run peek on each pane in sequence:

peek api
peek tests
peek refactor

Each response arrives as a separate Telegram message. Scan them on your phone screen to confirm all three agents are making progress.

Error Alerts: Know Before You Check

Remocode monitors terminal output across 30+ error patterns, including compilation failures, runtime exceptions, test failures, permission errors, and network timeouts. When an agent hits one of these, you get an immediate Telegram alert.

This means you do not need to peek constantly. If your phone is silent, your agents are running cleanly. If it buzzes, something needs attention.

What Error Alerts Look Like

An alert message tells you which pane triggered it and shows the relevant terminal output. You see the error context without needing to read the full log.

This runs at zero AI cost. The pattern matching happens locally inside Remocode with no API calls.

Taking Action with Via Mode

When an error alert arrives and you need to intervene, switch to via mode:

via api

Now every message you type in Telegram goes directly to the api pane as terminal input. You can answer questions, approve tool calls, or type commands. When you are done, type exit to leave via mode.

Approving Prompts Remotely

If Claude Code is waiting for approval on a file edit, you will see the prompt through via mode. Type 1 to approve or use the inline Telegram buttons that Remocode generates for numbered menus.

The AI Supervisor: Hands-Free Monitoring

For true hands-free operation, enable the AI Supervisor on each pane before you leave. The supervisor scans the terminal every two seconds and uses your configured AI provider to decide what to do.

Write a brief project description so the supervisor understands the context. For example: "Building a REST API for user management. Approve file writes and test runs. Reject anything that modifies the database schema."

Three Levels of Autonomy

• ●AI Supervisor — reads your project brief, makes smart decisions, escalates when uncertain
• ●Auto-Yes — approves simple yes/no prompts with zero AI cost
• ●Manual via Telegram — you approve everything yourself through via mode

Mix and match across panes. Your stable refactor pane might use Auto-Yes while your new API pane uses the supervisor with a strict brief.

Standup Reports: Async Progress Updates

If you do not want to check manually at all, configure standup reports. Remocode uses AI to analyze terminal activity and generates a summary of what each agent accomplished, what it is working on, and whether any issues arose.

Schedule these for whatever cadence makes sense. A report every two hours while you are away gives you a clear picture of progress without any manual checks.

A Real Workflow Example

Here is what monitoring looks like during a two-hour lunch:

• ●12:00 — Leave desk with three agents running, supervisor enabled
• ●12:15 — Error alert: test agent hit a missing dependency. You via tests and type the install command
• ●12:45 — Peek at all three panes. API agent is 60% through the endpoints. Refactor agent finished
• ●1:15 — Standup report arrives. API agent completed all endpoints. Test agent recovered and wrote 14 tests
• ●1:30 — Return to desk. Review the work, commit, and move on

Your agents worked for 90 minutes straight. You spent about three minutes on your phone total. That is the power of remote monitoring done right.