Agent Code – open-source Mac app for managing AI coding agents
Agent Deck is a native macOS app for the Pi AI coding agent, providing a unified interface for managing agents, skills, prompts, sessions, and more. It features explicit scope visibility, parallel work with git branching, GitHub integration, subagents, memory, and on-device automations via Apple Foundation Models.
Notifications You must be signed in to change notification settings
Fork 2
Star 16
BranchesTags
Open more actions menu
Folders and files
NameName
Last commit message
Last commit date
Latest commit
History
12 Commits
12 Commits
.github
.github
agent-deck-documentation
agent-deck-documentation
agent-deck.xcodeproj
agent-deck.xcodeproj
agent-deck
agent-deck
agent-deckTests
agent-deckTests
docs
docs
scripts
scripts
validation-tools
validation-tools
.gitignore
.gitignore
AGENTS.md
AGENTS.md
LICENSE
LICENSE
README.md
README.md
Repository files navigation
Pi is a powerful coding agent. It also lives in a terminal. That means scrollback for transcripts, dotfiles for configuration, and copy-paste between issues, repos, and sessions. Agent Deck turns Pi into a native macOS workspace for agentic development: every agent, skill, prompt, command, and session in one window, with explicit scope, live structured output, and the project context you'd otherwise rebuild from memory every time.
Agent Deck does not replace Pi or embed its own agent runtime. It launches the installed pi CLI in JSONL RPC mode, manages the surrounding resources and UI, and passes Pi exactly the flags it needs. The result is both a native control surface for Pi sessions and a platform for organizing the agents, skills, prompts, and workflows you run through it.
Install
Download the latest signed and notarized .dmg from Releases and drag Agent Deck to /Applications. Updates ship through Sparkle — you'll see a native macOS update dialog when a new version is available.
Requires macOS 26 (Tahoe) and Apple Silicon. The Pi CLI must be installed and discoverable — Agent Deck's built-in Doctor walks you through it on first launch.
Build from source
git clone https://github.com/a-streetcoder/agent-deck.git cd agent-deck open agent-deck.xcodeproj
Then build the agent-deck scheme from Xcode 26.4+. Or from the command line:
xcodebuild -project agent-deck.xcodeproj -scheme agent-deck \ -configuration Debug -destination 'platform=macOS' \ CODE_SIGNING_ALLOWED=NO build
Design guarantees
Scope is always visible. Every agent, skill, and prompt shows where it comes from — Builtin, Global, Library, or Project — with a colored chip, an icon, and text. No ambient discovery surprises, no "where did this come from" mystery sessions.
Assignment is explicit. Agent Deck launches Pi with --no-skills, --no-extensions, --no-prompt-templates, --no-themes and selectively re-enables only what you assigned. A skill being on disk doesn't mean it's loaded. A skill being assigned does.
Built-ins are read-only. When you customize a bundled agent, Agent Deck writes an override file. The original is never modified. You can disable overrides and snap back to the bundled defaults at any time.
Writes are visible. Every action that modifies a file shows you what it will write and where. Report-only subagents produce artifacts in their own directory, not edits to your project.
Native through and through. SwiftUI, multi-window, keyboard shortcuts for everything, signed and notarized. No Electron, no web views faking it.
Sessions and transcripts
Streaming transcript with steering messages, thinking blocks, tool calls, plans, inline diffs, file previews, and color-coded status — filterable so you see exactly the detail you want.
Live plan checklist tracks the agent's todo/in-progress/done/blocked/skipped state as it works.
Rich composer with paste handling, @-file suggestions, macOS Dictation, and attachments (files, folders, images).
Auto-titling via Apple's on-device Foundation Model (or any model you pick) — no more "Session 47".
Idle parking frees system resources when you walk away.
Terminal handoff when you need raw CLI for one-off debugging.
Parallel work without merge headaches
Every new session can spin up its own git branch and isolated worktree under Application Support. Run three agents on three features at the same time — no stepped-on files, no conflicting commits. A dedicated Merge toolbar action lands the work back to the source branch, with configurable keep-or-discard for the worktree and branch.
GitHub, end to end
Issue board with Open/Closed columns, sub-issue progress, dependency tracking, and cross-repo search.
Issue-to-session in one click: title, body, labels, and comments are loaded as context, and a session starts pre-configured to work on it.
GitHub auth via the gh CLI or native OAuth, with connection status and avatar in the toolbar.
Subagents
Parent sessions stay orchestration-first. They delegate scoped work — exploration, planning, implementation, review — to native subagents the app launches and tracks directly.
Bundled starter pack: explorer, planner, coder, reviewer. Override or replace any of them.
Summary cards in the transcript with per-agent status, tokens, and duration.
Supervisor request cards render native macOS decision UIs when a child needs human guidance.
Worktree isolation for write-capable subagents — multiple writers won't clobber each other.
Parallel and chained graphs via managed_subagent, managed_parallel, and managed_chain.
Agents, skills, prompts, and commands
Agents, skills, prompts, and slash commands — all in a sidebar, all browsable, all toggleable. Build them once, then assign them per-project or globally.
Skills import from any folder, GitHub repo, or skills.sh URL via a blobless sparse clone. Agent Deck writes AI-generated summaries on import so you know what unfamiliar skills actually do before you enable them. Sync upstream and resolve conflicts with a Keep Mine / Take Remote sheet.
Prompts are reusable starting points; pick one and a session opens with it preloaded.
Agents carry name, description, system prompt overrides, tool restrictions, model overrides, thinking level, and a generated avatar. Importable and exportable.
Commands — bundled and user-imported TypeScript slash commands injected into sessions.
Agents that remember
Per-project memory captures decisions, runbooks, architecture, and prior failures — written by the agent during sessions via the agent_deck_memory_write tool, stored as Markdown, and injected back into future sessions within a budget you control. Outdated memories get marked stale, not deleted; you stay in audit.
Secret-scanning blocks memory writes that look like private keys, GitHub tokens, AWS keys, or password=/token=/secret= assignments.
Automations powered by Apple Foundation Models
The boring parts done locally and for free:
Session titles drafted as you go.
Commit messages generated from staged diffs, with a one-click Commit / Push / Commit & Push toolbar.
Avatar prompts for the Image Playground.
Skill summaries during import.
Each automation has its own model picker — use the on-device Foundation Model where it makes sense, drop down to a cloud model when you need more.
Models, providers, environment
Auto-discover models from configured Pi providers. Group by provider. Set defaults, per-agent, and per-session overrides. Hide noisy unused entries. Opt eligible OpenAI models into priority service tier with a bundled extension.
The Environment view manages .env files across scopes with secret masking — never modifying bundled resources.
Health and setup
The Doctor runs health checks on the Pi CLI, version, path resolution, and required env keys, with auto-fix suggestions. A 6-page welcome tour and a step-by-step setup wizard get you from zero to first session in minutes.
Screenshots
Models and providers
Skills and subagent assignment
Agent definitions
Documentation
In-depth docs live in agent-deck-documentation/:
System prompt logic — how Agent Deck's launch flags compose with Pi's prompt assembly.
Pi RPC launch flags — the full surface of subprocess context.
Skills and model & thinking reference.
Memory design.
Resource refresh and file watching.
Conceptual docs under concepts/, reference/, and contributors/.
Contributor invariants and UI conventions live in docs/agent-guidelines/.
Requirements
macOS 26 (Tahoe) on Apple Silicon
Xcode 26.4+ for building from source
A working install of the Pi CLI
License
MIT License. See LICENSE.
About
Agent Deck
agentdeck.site
Topics
skills
models
pi
agents
prompts
llm
llms
llm-agents
Resources
Readme
License
MIT license
Uh oh!
There was an error while loading. Please reload this page.
Activity
Custom properties
Stars
16 stars
Watchers
1 watching
Forks
2 forks
Report repository
Releases 8
Agent Deck 1.7
Latest
May 31, 2026
+ 7 releases
Packages 0
Uh oh!
There was an error while loading. Please reload this page.
Contributors
Uh oh!
There was an error while loading. Please reload this page.
Languages
Swift 99.5%
Other 0.5%