Far lighter
About 150 MB after launch, down from about 1.25 GB.
NewAgentDeck v0.44 · lighter, with security skills
Eight agents running. Which one needs you?
Run Claude, Codex, Copilot, Gemini and the rest as a team of real terminal panes — then see which agent is working, idle or waiting for you.
Windows 10 / 11 · 64-bit · Free tier, no card
Not code-signed yet — if SmartScreen says “Windows protected your PC”, click More info → Run anyway. Why this happens
Interactive previewClick anything · type in a pane
Choose a review, or type a prompt into any pane.✳write a test for auth.ts⠋Working1
✳ Claude Code v2.1.0
E:\Projects\api
> write a test for auth.ts
⏺ Reading src/auth.ts
⠋ Working… 21s · ↑3.5k · esc to interrupt
⌘Codex · new paneIdle2
⌘ Codex v0.158.0
E:\Projects\api
Type a prompt below and press Enteror choose a review above ↑
✦refactor payment.ts⠋Working3
✦ Gemini CLI v0.40.0
E:\Projects\api
> refactor payment.ts
⏺ Mapping call sites
⠋ Working… 39s · ↑5.5k · esc to interrupt
✳add zod validationNeeds you4
✳ Claude Code v2.1.0
E:\Projects\api
> add zod validation
⏺ Reading src/routes.ts
Permission required
Routine run history
Every run tracked, retried in a click, and caught up after downtime.
Security skills
28 ready-made skills, now with secrets, auth and CI audits.
25 colour schemes
New: Night Owl, Poimandres, Horizon, Moonlight, Cobalt2, Cyberpunk, Matrix.
A command center for the agents already on your machine
Every shell is a real terminal. Every pane knows which agent is inside it. AgentDeck keeps the team visible, the work isolated and the next decision one click away.
Every agent, sorted by who needs you
The Agents dashboard collects every pane across every workspace into Needs you, Working and Idle — then jumps straight to the one asking for input.
One command palette for the whole deck
Press Ctrl+Shift+P to find any command, open pane, workspace, agent, layout, shell, colour scheme or page without digging through menus.
Claude and Codex usage at a glance
Subscription usage lives beside the agents it affects, so you can see what is left in the current Claude and Codex windows before assigning the next task.
The right action, right on the pane
Rename a pane, restart its agent, hand off its conversation, move it to another workspace or isolate it in a fresh worktree from one focused menu.
Real multi-terminal workspaces
Every shell is a true terminal pane. Split panes, drag dividers, move work between workspaces — and the exact arrangement comes back after a restart.
Isolated git worktrees
Give every pane its own worktree so parallel agents never collide. Review each diff, then merge, open a PR or discard — all from the Worktrees page.
Conversation handoff
The ⤳ button on any pane carries the running conversation into a fresh one — resume the same agent with its full history, or hand the whole transcript to a different agent or model.
Routines on a schedule
Schedule an agent prompt — “every weekday at 9, review my open PRs” — and AgentDeck opens the agent and sends it at that time. Run history shows how every run went, and runs missed while the app was closed can catch up.
Skills & a skill library
Reusable SKILL.md instructions wired into every agent. Start from 28 ready-made skills — including a Security set for secrets, auth, threat models and CI — or let an agent improve your own.
Eight plugins
Connect GitHub, GitLab, Vercel, Jira, Linear, Supabase, Google Drive or LinkedIn once and the agent in every pane gets real tools for it. Tokens stay on your machine.
Voice-to-text input
Press Ctrl+Shift+X and talk. A floating voice strip transcribes speech on-device with whisper.cpp and types it at the active prompt.
Folders at your fingertips
Pin and search project folders in the sidebar, drag one onto any pane to type its path, preview files with a double-click, and cut, copy and paste like in Explorer.
A browser beside your agents
Open localhost or any URL in a built-in browser that shares the Agents panel’s space. Dev servers your panes start show up as one-click buttons, and Ctrl+click on a terminal link opens it there.
Built for directing, not tab hunting
Real screens from the current app. Follow the trail from the agent that needs you to the command or pane action that moves the work forward.
Connect a tool once. Every agent can use it.
A plugin links AgentDeck to a service you already use and gives the coding agent in each pane real tools for it — no pasted tokens, no copied commands. Eight are live today.
- 1
Connect once
Open the Plugins tab in AgentDeck and connect the service. GitHub uses a one-time device code; Vercel, Jira, GitLab, Linear and Supabase hand off to the agent’s own secure sign-in; Google Drive and LinkedIn use an OAuth app from your own developer account.
- 2
Pick what it can do
Read and review are on by default. Every write action is opt-in and asks before it runs.
- 3
Agents get the tools
Open a workspace, restart the agent once (↻), and the tools are already wired in — no pasted tokens, no copied commands.
GitHub
Version controlReview pull requests, read diffs and checks, triage issues and dispatch Actions — straight from a pane.
One-time device code
Vercel
Deploys & hostingInspect deployments, read build and runtime logs, and check projects and domains while the agent works.
Sign in from the pane
Jira & Confluence
Project managementRead and update Jira issues, search boards with JQL and pull Confluence pages — one connection covers the whole Atlassian suite.
Sign in from the pane
GitLab
Version controlWork merge requests, inspect CI/CD pipelines and triage issues across your GitLab projects — straight from a pane.
Sign in from the pane
Linear
Project managementFind, create and update Linear issues, projects and cycles while the agent works — no context-switch to the browser.
Sign in from the pane
Supabase
DatabasePoint an agent at your own Supabase project and ask it to review the database — schema, policies, migrations and logs, without pasting SQL or screenshots.
Sign in from the pane
Google Drive
FilesSearch your Drive and pull a document straight into the pane — specs, notes and briefs the agent should be working from.
Your own Google OAuth client
Let an agent run your job hunt — file postings, score them against your CV, draft tailored applications and track every one. Pair it with a routine and it runs every morning.
Your own LinkedIn app
BitbucketVersion controlComing soon
SentryMonitoringComing soon
NetlifyDeploysComing soon
Plugins are a Pro feature. GitHub’s and LinkedIn’s tools wire into most MCP-capable agents you have installed; Vercel, Jira, GitLab, Linear and Supabase reach Claude Code and opencode today; Google Drive is Claude Code only.
Start free. Upgrade when the deck fills up.
One flat price for unlimited agents, panes and workspaces — no per-agent fees. Bring your own Claude, Codex, Copilot or Gemini subscription.
Free
Kick the tyres on a real project.
Free
Forever. No card needed.
Download free- 1 workspace
- Up to 4 terminal panes
- All 12 coding agents
- Grid / column / row layouts
- Manual updates
- Community support
Most popular
Pro
The daily driver for solo builders.
Tk 749/mo
Tk 7,490/yr if paid yearly
Get Pro- Unlimited workspaces & panes
- Isolated git worktree per pane + Review/Merge
- Conversation handoff between agents (⤳)
- Routines, skills & all eight plugins
- Voice-to-text input (Ctrl+Shift+X)
- One-click background auto-updates
- Cloud settings, skills & profile sync
- Priority email support
Coming soon
Team
For squads running agents in parallel.
Coming soon
Pricing to be announced.
Notify me- Everything in Pro
- Centralised billing & seats
- Shared workspace templates
- SSO & guided onboarding
- Dedicated support channel
Prices in Bangladeshi Taka (Tk). You still pay separately for whatever coding agent or API you run inside AgentDeck.
From empty folder to a full deck in three steps
AgentDeck, answered
Plugins connect AgentDeck to a service you already use and give the agent in each pane real tools for it. Eight are live: GitHub (review pull requests, read diffs and checks, triage issues, run Actions), Vercel (inspect deployments and read build and runtime logs), Jira & Confluence (search boards, read and update issues, pull pages), GitLab (merge requests, pipelines and issues), Linear (find, create and update issues and projects), Supabase (review your database read-only — schema, RLS policies, migrations and logs) and Google Drive (search your Drive and read documents into the pane) and LinkedIn (a job-hunt pipeline that scores postings against your CV and drafts applications — it never applies for you). GitHub’s token and the Google and LinkedIn client secrets stay encrypted on your machine; Vercel, Jira, GitLab, Linear and Supabase sign in through the agent so AgentDeck never sees a token. Every write capability is opt-in, and plugins are a Pro feature. See the Plugins page for the full ability list.
Yes. The ⤳ button on any pane hands the running conversation into a new pane in the same workspace. Pick the same agent and it resumes its own session with the full history (Claude Code, Codex, opencode, Goose); pick a different agent or model and AgentDeck exports the whole conversation to a Markdown transcript and starts the new agent on it. Conversation handoff is a Pro feature.
Yes. Open a workspace with “Isolate each terminal in its own git worktree” ticked and every pane gets its own worktree on a throwaway branch, so agents never overwrite each other or your checkout. The Worktrees page shows each diff with Merge to base, Open PR and Discard. Isolated worktrees are a Pro feature.
Yes — that is what Routines are. Give one a prompt, an agent, a time and the days, and AgentDeck opens the agent and sends the prompt on schedule while the app is running. Pair them with Skills (reusable instructions every agent follows) and plugins for hands-off jobs like a morning PR review. Routines and Skills are Pro features.
There is a free tier — one workspace and up to four panes, with every one of the 12 coding agents. Pro (Tk 749/mo, or Tk 7,490/yr) unlocks unlimited workspaces and panes, isolated worktrees, conversation handoff, routines, skills, plugins, voice input, one-click auto-updates and settings sync. You still pay separately for whatever coding agent or API you run inside it.
Windows 10/11 (64-bit), macOS 12 or later on Apple Silicon, and Linux (64-bit AppImage). The core experience works across all three. Voice dictation is available on Windows and macOS; Linux is the exception for now. The macOS build is not signed with an Apple Developer ID yet, so first launch requires Open Anyway in Privacy & Security.
No. Install AgentDeck, and for any agent you have not set up yet it shows the exact install command plus a link to the guide, then re-checks your PATH so you can start without restarting.
The installer sets AgentDeck up per-user (no admin prompt) and enables one-click self-updates. The portable ZIP runs from any folder with no install, but you update it manually. On a Mac, open AgentDeck-osx-Setup.pkg, or unzip the Mac app ZIP and drag AgentDeck.app into Applications. On Linux, download AgentDeck-Linux-Install.sh and run `bash AgentDeck-Linux-Install.sh` — it unpacks the app and adds it to your app launcher, no chmod needed. Prefer a plain portable executable instead? The bare AppImage is available too — `chmod +x AgentDeck.AppImage && ./AgentDeck.AppImage` runs it directly with no install step.
Yes, for now. The Mac build isn’t signed with an Apple Developer ID yet. Click Done, then open System Settings → Privacy & Security and click Open Anyway. You only need to do it once.
Yes, for now. The builds are not code-signed yet, so SmartScreen shows “Windows protected your PC”. Click “More info” → “Run anyway”.
Nowhere by default. AgentDeck is a local app — terminals, agents and keys never leave your machine. An optional account only syncs your app preferences.
Run your first deck of agents
Windows 10 / 11 · macOS 12+ (Apple Silicon) · Linux 64-bit · free tier
Windows 10 / 11 · 64-bit · Free tier, no card
Not code-signed yet — if SmartScreen says “Windows protected your PC”, click More info → Run anyway. Why this happens