Running Claude Code on Ten Projects at Once Without Losing Your Mind
You've got Claude Code running in one terminal, OpenCode in another, and something else chugging along in a third. Now you're alt-tabbing between windows trying to remember which session was waiting on your input and which one finished ten minutes ago. Agent Deck is a terminal-based mission control for exactly that problem.
What It Does
Agent Deck is a command center for managing multiple AI coding agents from a single terminal window. Instead of juggling separate terminal sessions, you get one view that shows every running agent — whether it's actively working, waiting for input, or done. A single keystroke switches between sessions.
The tool is written in Go (version 1.25.13) and runs on macOS, Linux, and Windows via WSL. It's MIT-licensed and installs with a one-line curl command. Beyond the basic session management, it layers on grouping, search, session forking, git worktree integration, and cost tracking. There's also a "conductor" feature that lets you control things from your phone, which the README highlights as a way to keep a whole fleet of agents manageable.
The mental model here is straightforward: you're not replacing your agents, you're organizing them. Each agent still runs its own process in its own context, but Agent Deck gives you a unified interface for seeing what's happening and jumping between tasks.
Why It's Cool
The problem is real and the solution is focused. If you've ever run three coding agents simultaneously, you know the pain isn't the agents themselves — it's the context switching overhead. Agent Deck doesn't try to be an AI orchestrator or a workflow automation platform. It's a dashboard. That restraint is what makes it useful.
Groups and search scale with your workload. When you're running five agents, you can probably keep track mentally. At fifteen, you can't. Being able to group sessions and search across them means the tool stays useful as your usage grows, rather than becoming another thing you have to manage.
Git worktrees plus forking is a smart combination. If you're using agents to explore different implementation approaches, you can fork a session and use git worktrees to keep the branches isolated. That's a workflow that makes sense for how people actually use coding agents — try something, branch off, try another angle — rather than forcing you into a linear process.
Cost tracking is the feature you don't think you need until you do. Running multiple agents across multiple projects adds up. Having visibility into what each session costs means fewer surprises and better decisions about when to let an agent keep grinding versus when to step in.
The contributor pipeline is unusually transparent. The README states that every incoming PR is validated — applied, built, tested — within about a day, and good ones merge in the next release batch. Recent releases have shipped dozens of community fixes. That's a concrete commitment, not a vague "we welcome contributions" boilerplate. There's even a machine-readable intake spec for contributors who want to use an AI agent to help write the PR, which is a nice bit of dogfooding.
They're actively looking for co-maintainers. Beyond one-off PRs, the project wants one or two regular co-maintainers to own specific areas — a tool integration, the TUI, the web view, CI. If you've been looking for a project to get more deeply involved with, this is an explicit invitation.
How to Try It
Getting started is a single command:
curl -fsSL https://raw.githubusercontent.com/asheshgoplani/agent-deck/main/install.sh | bash
Then run:
agent-deck
That's it. The README doesn't include a lot of configuration detail in the portion available, so once you're running, you'll want to explore the tool's own interface and documentation for specifics on setting up groups, configuring cost tracking, or connecting the conductor.
If you want to contribute, the README lays out a clear path:
- Read CONTRIBUTING.md for how the validation pipeline works and what makes a PR land fast.
- If you're contributing with an AI agent, point it at .github/INTAKE.md or the agent-deck-contributor skill.
- Comment on an issue before starting so you don't duplicate work.
- Look for issues labeled
good first issue.
There's also a Discord if you want to ask questions or coordinate.
You can find the full repository at github.com/asheshgoplani/agent-deck.
Final Thoughts
Agent Deck is best suited for developers who are already running multiple AI coding agents regularly and feeling the friction of managing them across separate terminal windows. If you're running one agent on one project, this is probably overkill. But if you're at the point where you've got a mental list of which session is where and what state it's in, the value proposition is clear.
The project is actively maintained, the contribution process is well-documented, and the feature set addresses a real workflow problem without overreaching. It's worth a look if you're managing agent sprawl — and if you've been wanting to get more involved in an open-source project, the co-maintainer invitation is unusually direct.
Follow @githubprojects for more developer tools and open source projects.