One terminal cockpit for Claude, Gemini and other agents
Agent Deck is a Go-based TUI that keeps sessions for Claude Code, Gemini CLI and six other agents in one place, with a cost dashboard.
Symbolic image: a workstation with several split terminal panes, blinking status lights and hands at the keyboard.
Agent Deck is an open-source terminal program that gathers the running sessions of eight AI coding agents into a single interface.
At a glance
- Written in Go and published under the MIT license; the repository lists 847 stars, 147 forks and 2,951 commits on main.
- Claude Code and Gemini CLI are fully integrated; six further tools get status tracking and grouping only.
- The built-in cost dashboard ships prices for 15 models.
- A web dashboard runs at 127.0.0.1:8420 by default, and sessions can be isolated inside Docker containers.
- The project claims a shared socket pool cuts MCP memory use by 85 to 90 percent.
Agent Deck is an open-source terminal program that gathers the running sessions of eight AI coding agents into a single interface. Anyone driving two or three assistants at once usually ends up with a wall of terminal windows and no quick way to tell which session is still working and which one has been sitting idle waiting for input.
Two tiers of support
The project is explicit that support comes in two depths. Claude Code and Gemini CLI are wired in fully. Six more tools — OpenCode, Codex, Copilot, Cursor in the terminal, Crush and the DeepSeek harness — are tracked, labeled with a status and sorted into groups, but without the deeper hooks. Additional command-line tools can be registered through a configuration file.
What the interface does
The text interface is built on the Bubble Tea framework and drives tmux underneath. Every session carries one of four states: running, waiting, idle or error. Fuzzy search, single-key shortcuts for creating and forking sessions, and a per-session manager for MCP servers round out the main view, so attaching a server does not mean hand-editing config files.
A browser dashboard at 127.0.0.1:8420 mirrors the same fleet. Sessions can be forked, archived and bound to git worktrees, which keeps parallel branches of the same codebase from stepping on each other.
Spend and memory
A separate dashboard tracks token usage and converts it into money using prices stored for 15 models, spanning Anthropic, Google and OpenAI systems. For larger fleets the project points to a shared socket pool that it says reduces MCP server memory use by 85 to 90 percent. Sessions can also run inside Docker containers with the code bind-mounted, and instances on remote machines can be driven over SSH from a local terminal.
What is not confirmed
Every figure here comes from the project's own description; no independent second source covers this release. The memory saving is a vendor claim that we did not measure. The 847 stars and 2,951 commits are a snapshot of the repository, not evidence of day-to-day adoption. Installation runs through a shell script, Homebrew or a prebuilt binary, and building from source is documented against Go 1.25.13.
FAQ
Which AI coding agents does Agent Deck support?
Claude Code and Gemini CLI are fully integrated. OpenCode, Codex, Copilot, Cursor in the terminal, Crush and the DeepSeek harness are tracked with status and grouping, and other tools can be added through a configuration file.
Is Agent Deck free?
The program is MIT-licensed and free to use. The cost comes from the models behind it, which is why the tool includes a dashboard that prices token usage across 15 models.
Does Agent Deck run on Windows?
The project lists macOS, Linux and WSL, so Windows is covered only through the Linux subsystem. Building from source is documented against Go 1.25.13.
