Image Agent Deck

Built on the CLI. A native Mac app for AI coding agents.

Agent Deck is the native macOS app for Pi CLI – a supercharged cockpit that turns terminal-based agent sessions into a visual, project-aware workflow. Author specialist AI coding agents, curate per-project skills, and run parallel sessions in isolated git worktrees, all from a native Mac interface that sits next to your editor, not inside it. Watch every plan, tool call, and diff stream live in a color-coded transcript. Launch a session from any GitHub issue, run it in an isolated worktree, and merge the result back when it's done.

Download for macOS GitHub 113
from the terminal
  • Free & open source
  • macOS 26+
Release
v5.1.3
What you get

The interface

Cockpit

A Mac app you can actually use.

Projects, sessions, and GitHub issues stay one click apart, not one terminal apart. You see what every session is doing, in real time, in a window that lives next to your editor.

  • Color-coded streaming transcript with plan, tool calls, and inline diffs
  • Project and session sidebar with multi-window, idle-parking, and terminal handoff
  • Built-in Doctor that diagnoses your Pi CLI setup
Branch

Fork the conversation, not your focus.

Any session can branch. Fork it to explore a different direction with the transcript carried over, retry a prompt in a fresh session, or open a 1:1 with a single agent on its own model. Your context never gets cornered.

  • Fork a session with the parent transcript captured as context
  • Retry the same prompt without rebuilding the conversation
  • 1:1 sessions with any agent, each with its own model and thinking level

Your agents

Library

A real library for what your agents use.

Cherry-pick individual skills from any GitHub repo or skills.sh URL. No need to clone the whole package. Keep upstream in sync. Decide which prompts, skills, and slash commands are active per project.

  • Import individual skills from GitHub or skills.sh URLs
  • Builtin, Global, Library, and Project scopes for what is active where
  • AI-generated summaries so you know what each skill does before enabling it
Prompts

Write it once, run it everywhere.

Saved prompts and slash commands live in the same library as skills. Ship with builtins like investigate-a-bug and plan-a-feature, add your own, and assign them per project so the right shortcuts are always one keystroke away.

  • Builtin prompts for everyday workflows, plus your own
  • Slash browser in the composer: commands, prompts, and skills in one menu
  • Per-project assignment, same as skills and agents
Author

Build agents, run them in parallel.

Agents are first-class objects with their own prompt, tools, skills, model, and avatar. Run them side by side in isolated worktrees, then merge completed work back with one click.

  • Custom prompt, tools, skills, model, and avatar per agent
  • Per-session git branch and worktree isolation
  • One-click merge of completed worktrees back to the source branch

Your workflow

Triage

Issues are the start, not a detour.

Browse repo issues inside a native board, pick one, and launch a session with the title, description, labels, and comments already loaded as context. Triage and execution stay connected.

  • Native GitHub issue board inside the app
  • Issue context auto-loaded into the session
  • Sub-issue progress tracked alongside execution
Memory

Your projects remember what matters.

Turn on memory per project and sessions quietly accumulate the preferences, decisions, and gotchas that usually get retyped. Memories are plain files you can read, edit, and delete, recalled on demand instead of stuffed into every prompt.

  • Per-project memory you can switch on or off
  • Every memory is a readable file with type, scope, and history
  • Recalled only when relevant, so it never bloats your context window
Autopilot

The chores run themselves.

Sessions name themselves with AI and keep their titles fresh as plans change. Commit and push from the toolbar with a model-written message. Let each new session run in its own branch and worktree, merged back when the work lands.

  • AI-generated session titles that update as the plan evolves
  • Commit and push toolbar actions with model-written messages
  • Optional worktree isolation: every session on its own git branch

Make it yours

Models

Bring your own keys. Pi handles the auth.

No extra subscription. Use your existing Claude, OpenAI, or local-model setup through Pi. Agent Deck adds a per-session picker, on-device drafting, and runtime tools.

  • Per-session model picker across every Pi provider
  • On-device Foundation Model for titles, commit messages, and summaries
  • OpenAI fast mode and Exa web search built in
Yours

Make it yours.

Themes restyle the whole canvas, not just the accent color. Pick a preset like Tokyo Night, Nord, or Catppuccin Mocha, or build your own palette, and the entire app follows, from the sidebar to the transcript.

  • Eleven built-in themes, fully custom palettes supported
  • Background, surfaces, and accents all follow the theme
  • Transcript display options: decide exactly what you see

Why we built Agent Deck

AI coding agents are powerful – but running them from a terminal means context switching, lost transcripts, and manual git hygiene. Existing AI code editors bundle the agent inside the editor, forcing you into one workflow.

We wanted something different: a dedicated cockpit that lives next to your editor, not inside it. A place where you author specialist agents per project, run them in parallel in isolated git worktrees, and merge the results when you're ready – without ever touching your working tree.

Agent Deck is built on Pi CLI, the open-source agent harness. It is free, open source, and built by developers who use it daily.

How Agent Deck compares

Agent Deck is the only native Mac app with a per-project skills library, GitHub issue integration, and per-project memory – features no other coding agent GUI offers.

Capability Agent Deck T3 Code Synara Osaurus CLI alone
Native Mac app ✓ SwiftUI Electron Electron ✓ Swift Terminal only
Parallel worktree sessions ✓ Isolated branches One-at-a-time ✓ Isolated branches Not supported Manual setup
Per-project skills library ✓ Cherry-pick from repos Not available Not available Not available Manual config
GitHub issue → session ✓ Native board Not available Not available Not available Copy-paste
Per-project memory ✓ Auto-accumulates Not available Not available Global only Not built-in
Multi-agent support Pi CLI agents ✓ 4 agents ✓ 8+ agents ✓ Any model Single agent
One-click PR ✓ With AI message ✓ Auto-generated ✓ Auto-generated Not available Manual

Frequently asked questions

What makes Agent Deck different from AI code editors like Cursor?
Agent Deck is not an editor – it is a dedicated cockpit for AI coding agents that sits next to your existing tools. You bring your own editor, API keys, and models through Pi. It specialises in parallel agent sessions in isolated git worktrees, per-project skill libraries, and launching sessions directly from GitHub issues.
How does Agent Deck compare to T3 Code, Synara, or Osaurus?
Agent Deck is the only native Mac app with a per-project skills library, GitHub issue integration, and per-project memory – features no other coding agent GUI offers. For a full side-by-side breakdown, see the comparison page.
Is Agent Deck really free?
Yes. Agent Deck is free and open source. You only pay for the AI model usage through your own API keys (Claude, OpenAI, or local models) – there is no Agent Deck subscription.
What macOS versions are supported?
Agent Deck requires macOS 26 (Tahoe) or later. It is a native Mac app built with SwiftUI and AppKit.
Do I need to install anything besides Agent Deck?
You need Pi CLI, the open-source agent harness that Agent Deck builds on. The install script sets up both in one command: curl -fsSL https://raw.githubusercontent.com/a-streetcoder/agent-deck/main/install.sh | bash.
Can I use Agent Deck with my existing projects?
Absolutely. Point Agent Deck at any local git repository and it will create isolated worktree branches for each session, never touching your working tree until you merge the results.
How does per-project memory work?
When enabled, Agent Deck stores preferences, decisions, and gotchas as plain Markdown files inside your project. They are recalled on demand – never automatically injected into every prompt – so your context window stays clean and relevant.
How does Agent Deck compare to running agents directly from the terminal?
The terminal gives you one session at a time with no visual transcript, no skill library, and manual git branch management. Agent Deck adds a color-coded streaming transcript, per-project skill curation, native GitHub issue integration, and automatic worktree isolation – all in a Mac-native interface. See the comparison table above for a side-by-side breakdown.
Which AI models can I use with Agent Deck?
Agent Deck works with any model supported by Pi CLI, including Claude (Anthropic), GPT (OpenAI), and local models via Ollama. You bring your own API keys – Agent Deck adds a per-session model picker, on-device drafting via the Foundation Model, and built-in Exa web search. There is no vendor lock-in.

Try Agent Deck

Free, open source, built for macOS 26 and later.

Download for macOS GitHub 113