Ashesh Goplani

Posted on Feb 3

How to Manage Multiple Claude Code, Gemini, and Codex Sessions in One Terminal

#claudecode #ai #codex #gemini

I usually have 5-10 AI coding agents running at the same time. Claude Code on the frontend, another Claude on the backend, Gemini researching architecture decisions, Codex writing tests. At some point, managing terminal tabs became the bottleneck, not the coding itself.

So I built Agent Deck, an open-source terminal session manager that gives me one dashboard for all of them.

The Problem With Multiple AI Sessions

If you use more than one AI coding agent, you know the pain:

Too many terminal tabs. Which tab was Claude? Which was Gemini? Was Codex even still running?
No visibility. Which agent is working? Which one finished and needs your input? You have to check each one manually.
Context switching. Jumping between tabs, scrolling up to remember what you asked each agent, losing your train of thought.

I tried tmux sessions, multiple terminal windows, split panes. None of them gave me what I actually needed: a single view of what every agent is doing right now.

How I Set It Up

Install

curl -fsSL https://raw.githubusercontent.com/asheshgoplani/agent-deck/main/install.sh | bash

# Or: brew install asheshgoplani/tap/agent-deck
# Or: go install github.com/asheshgoplani/agent-deck/cmd/agent-deck@latest

Create Sessions

I create sessions for whatever I'm working on that day:

agent-deck add ~/my-project -t "frontend" -g "web-app" -c claude
agent-deck add ~/my-project -t "backend" -g "web-app" -c claude
agent-deck add ~/my-project -t "research" -g "web-app" -c gemini
agent-deck add ~/my-project -t "tests" -g "web-app" -c codex
agent-deck add ~/infra -t "refactor" -g "infra" -c opencode

Each session gets a title, a group for organization, and a tool. Or just press n in the TUI to create sessions interactively.

How Status Detection Works

This is the core feature. Agent Deck polls each session every 2 seconds and shows you what every agent is doing:

Status	Symbol	Color	What It Means
Running	`●`	Green	Agent is actively generating output
Waiting	`◐`	Yellow	Agent finished and needs your input
Idle	`○`	Gray	You've seen it, nothing to do
Error	`✕`	Red	Session crashed or doesn't exist

When I come back from a break, I instantly see which agents need attention. No scrolling through tabs.

This works for Claude Code, Gemini CLI, Codex, and OpenCode out of the box. Custom tools can define their own detection patterns in config.toml.

How I Switch Between Sessions

From the Dashboard:

j/k or arrow keys to navigate
Enter to attach to a session
Ctrl+Q to detach back to the dashboard

Switching is instant because everything runs in tmux underneath.

Quick Filters:

When I have 10+ sessions, I use status filters:

@ — show only sessions waiting for input
! — show only running sessions
/ — fuzzy search across all session names

I press @, work through the waiting sessions one by one, then go back to the full list.

Notification Bar:

Waiting sessions also show up in the tmux status bar:

⚡ [1] frontend [2] api [3] backend

Press Ctrl+b 1 to jump directly to any waiting session without even opening the TUI.

Managing MCP Servers Without Editing JSON

I have ~25 MCP servers defined (web search, Playwright, GitHub, filesystem, etc.). Instead of editing .mcp.json by hand every time I need one, I press M in any session:

Space to toggle an MCP on/off
Tab to choose scope: LOCAL (this project only) or GLOBAL (all projects)
Enter to apply

The session restarts automatically with the new MCPs. Define your MCPs once in ~/.agent-deck/config.toml:

[mcps.playwright]
command = "npx"
args = ["-y", "@playwright/mcp@latest"]

[mcps.filesystem]
command = "npx"
args = ["-y", "@modelcontextprotocol/server-filesystem", "/Users/me"]

Then toggle them per session as needed.

Consulting Multiple AI Models From One Conversation

This is my favorite workflow. Sometimes I want opinions from different models before making an architecture decision. Using the Agent Deck skill for Claude Code, I can say:

"Consult with Gemini and Codex: what's the best approach for real-time updates in a Go backend?"

Claude automatically:

Spawns a Gemini session and a Codex session
Sends both the same question
Waits for their responses
Pulls the outputs back and compares all three perspectives

Three AI opinions from a single conversation. The sessions are visible in the TUI, and everything is managed through CLI commands.

Install the skill:

/plugin marketplace add asheshgoplani/agent-deck
/plugin install agent-deck@agent-deck

Forking Claude Sessions

Press f to fork any Claude Code session. The fork inherits the full conversation history, so I can try a different approach without losing my original context.

f — quick fork with auto-generated name
F — fork dialog with custom name, group, and optional git worktree isolation

If the fork works out, great. If not, the original is untouched.

Agent Deck vs Other Approaches

Approach	Pros	Cons
Multiple terminal tabs	Simple, no setup	No status visibility, impossible with 5+ sessions
tmux manually	Persistent sessions	No AI-specific features, manual status checking
API-based orchestration	Full control	Requires coding, no interactive sessions
Agent Deck	Status detection, MCP management, forking, multi-tool	Requires tmux

Supported Tools

Tool	Status Detection	MCP Management	Fork	Resume
Claude Code	Yes	Yes	Yes	Yes
Gemini CLI	Yes	Yes	No	Yes
Codex	Yes	No	No	No
OpenCode	Yes	No	No	No
Custom CLI tools	Configurable	No	No	No

Quick Start Cheatsheet

# Install and launch
curl -fsSL https://raw.githubusercontent.com/asheshgoplani/agent-deck/main/install.sh | bash
agent-deck

# Key shortcuts
# n       — New session
# Enter   — Attach to session
# Ctrl+Q  — Detach back to dashboard
# M       — MCP Manager
# f       — Fork session
# /       — Search
# @       — Filter waiting sessions
# ?       — Full help

What's Next

I'm working on a few things:

Automated issue triage: GitHub webhooks trigger agent sessions to analyze new issues
Hook-based workflows: Events (new PR, failed test) automatically launch the right agent
Session templates: Predefined session setups for common workflows

DEV Community