Using Via Mode With Gemini CLI: Remote Control Guide

Gemini CLI and Remote Terminal Control

Google's Gemini CLI brings Gemini's capabilities to the terminal for code generation, analysis, and refactoring. Like Claude Code, it is built on the Ink framework, which means Remocode's Ink-specific optimizations apply automatically. This guide covers the specifics of using via mode with Gemini CLI.

Automatic Ink Detection

When you connect to a terminal running Gemini CLI via via <name>, Remocode detects the Ink application and activates the same optimizations used for Claude Code:

• ●UI chrome removal — Gemini CLI's interface elements, borders, and formatting decorations are stripped from the output sent to Telegram. You receive clean, readable text.
• ●ANSI code stripping — Color codes, cursor positioning, and other escape sequences are removed.
• ●Smart typing — Text is sent first, followed by Enter after a 1-second delay to allow the Ink renderer to process input.
• ●Graduated retry — If no output appears after input, retries occur at 8, 18, and 28 seconds.

Gemini CLI's Interaction Style

Gemini CLI has its own patterns for requesting user input. It may ask for confirmation before applying changes, request clarification on ambiguous instructions, or present options for different approaches. These patterns are handled by both Remocode's question detection (when you are not in via mode) and directly through via mode text input.

Starting a Gemini CLI Via Session

Launch Gemini CLI in a named Remocode terminal, then connect from Telegram:

via gemini-session

You begin seeing Gemini's output as it processes your initial instructions. The output is polled every second, and only new content (deltas) are sent to keep your Telegram chat manageable.

Sending Instructions

Type your instructions naturally in the Telegram chat:

Generate a REST API for a todo application using Express.js with TypeScript

Remocode's smart typing sends this to Gemini CLI with proper Enter-key handling. The graduated retry system ensures delivery even if Gemini is mid-processing when your message arrives.

Handling Gemini's Responses

Gemini CLI typically generates longer responses that include code blocks, explanations, and file paths. In via mode, these arrive as delta messages in Telegram. Since the output is polled every second, a long Gemini response may arrive across several Telegram messages.

To read Gemini's complete responses more cleanly, use the escape command:

!reply_3

This retrieves the last three complete AI responses from the terminal buffer, formatted more coherently than the delta stream.

Status Reports for Gemini Sessions

Gemini CLI sessions can be long-running, especially for complex generation tasks. Use status to get AI-generated progress summaries:

!status

While in via mode, the status report appears in your Telegram chat alongside the streaming output. This gives you a high-level view of what Gemini has accomplished without needing to read all the individual delta messages.

Comparing Gemini CLI via Mode to Claude Code

Both are Ink applications, so the core via mode behavior is identical. The differences are in the agents themselves:

• ●Gemini CLI tends to produce more detailed explanations inline with code generation. Expect more verbose output in your via mode stream.
• ●Claude Code tends to ask more frequent permission questions. Via mode with Claude Code involves more back-and-forth input.
• ●Both benefit equally from smart typing and retry logic.

Practical Gemini CLI Workflow

• ●Start Gemini CLI in a terminal named "gemini" with a code generation task.
• ●Head to a coffee shop. Connect with via gemini.
• ●Gemini generates a file structure and asks: "Should I include unit tests?" Type "yes, use Jest."
• ●Gemini continues generating. You monitor the output, occasionally checking !status for summaries.
• ●Gemini asks about database choice. Type "PostgreSQL with Prisma ORM."
• ●Generation completes. You !exit via mode.
• ●Later, you run audit gemini to check the generated code for security issues before merging.

Tips for Gemini CLI Specifically

Gemini CLI can produce large amounts of output. If the delta messages in via mode are overwhelming, consider a workflow where you let Gemini work unmonitored and periodically check in with status and reply rather than maintaining a continuous via mode connection. Reserve via mode for the interactive phases where Gemini needs your input.

The combination of passive monitoring through question detection and alerts, plus active via mode for interactive moments, gives you the optimal balance of awareness and efficiency when working with Gemini CLI remotely.