Reply Command: Retrieving AI Agent Responses From Terminal History

Catching Up Without Scrolling

When you have been away from your terminal and an AI coding agent has been working, you return to a wall of output. Scrolling through it all on a phone screen is tedious. The reply command gives you a focused view of just the AI agent's responses.

Basic Usage

Send reply to your Remocode Telegram bot, and it retrieves the most recent AI response from the terminal buffer. This is the last substantive output from Claude Code, Gemini CLI, OpenAI Codex, or whatever agent is running.

To get multiple responses, use the numbered variant:

reply_5

This fetches the last 5 AI responses, giving you a chronological summary of what the agent has been doing.

What Counts as an AI Response

Remocode's buffer analysis distinguishes between different types of terminal output. AI responses are the substantive text blocks generated by the agent, as opposed to command prompts, status indicators, progress bars, and UI chrome. The reply command extracts these response blocks and presents them in order.

For Ink-based apps like Claude Code, this means you get the agent's actual explanations, code proposals, and questions — not the surrounding interface elements.

Reply vs. Via Mode

Reply and via mode serve different purposes:

• ●Reply is retrospective. It shows you what already happened. It is read-only and does not establish a streaming connection. Use it to catch up.
• ●Via mode is real-time. It shows you what is happening now and lets you interact. Use it when you need to participate in the session.

A common workflow: use reply_3 to understand the current state, then decide whether you need to enter via mode for hands-on interaction or if the agent is handling things fine on its own.

Using Reply in Via Mode

You can use reply while in via mode with the escape prefix:

!reply_5

This is useful when you just connected to a session and want context before the streaming output tells you enough. The reply results appear in your Telegram chat alongside the via mode output, giving you both historical and real-time views.

Choosing the Right Number

How many responses should you request? It depends on how long you have been away and how active the agent has been:

• ●`reply` (1 response) — Quick check. "What just happened?"
• ●`reply_3` — Short absence. "What did I miss in the last few minutes?"
• ●`reply_5` to `reply_10` — Longer absence. "Give me the full picture."

Requesting too many responses results in a long Telegram message that might be as overwhelming as the terminal output itself. Start with 3-5 and increase if you need more context.

Practical Example

You started Claude Code on a task to add user authentication to your API at 2 PM. At 3 PM, you check in:

reply_5

You receive five responses showing that Claude: (1) analyzed the existing codebase, (2) created a JWT middleware, (3) added login and registration endpoints, (4) wrote tests, and (5) is now asking whether to add refresh token support. You tap the Yes button on the forwarded question, and Claude continues.

The entire check-in took 20 seconds and you have full context on an hour of work.

Reply and Status

For an even higher-level overview, combine reply with status. The status command generates an AI-powered summary of the entire terminal session, while reply gives you the raw agent responses. Status tells you "what happened" in a paragraph; reply shows you "what was said" verbatim.