DeckSpacev3.7.2 · macOS, Windows, LinuxDeckSpace runs the agent CLIs already on your PATH, on your own accounts — eleven of them, plus any you register yourself. DeckSpace is not tied to any model vendor, and has no stake in which one you use. Every CLI is a peer: hand a task from one to another mid-flight, have one vendor’s model review another’s diff, and compare several attempts at the same job side by side before you keep one. Bring your own key — no metering, no markup, and your code never leaves your machine.
› worktree ready — pick a task below› awaiting taskEleven agent CLIs are built in, and DeckSpace detects whichever are already on your PATH. You sign in to each with your own account, and the work moves between them: hand a task from one to another when it runs out of context, and have one vendor’s model review a diff another vendor’s model wrote.
Independence is what makes the rest of it work. A task can move between CLIs because none of them is the house model, and a review means something because the reviewer was not the author. You are never choosing a vendor when you choose the deck.
Run the same job several times over — different models, different prompts, different approaches — each on its own git worktree, and read the results next to each other before anything touches your branch. Not old-versus-new. Old versus every new.
Generating code is no longer the slow part. Deciding which version to trust is — and that is the part this is built for.
Agents bill by the token, so the number that matters is the one that arrives with the invoice. DeckSpace picks a model to match the job rather than sending everything to the most expensive one, tells you which it picked and why, and stops at a ceiling you set.
We never meter you and never mark anything up. You are on your own accounts, at your own rates — we are the thing that stops those rates surprising you.
3.7.0 is the release that leaves macOS. Windows and Linux installers are published and served, not built and thrown away. Underneath, the deck stops making choices on your behalf: the assistant ships switched off, the routing mode is a question on first run, and a plugin you install once is reachable by every agent CLI you drive.
Signed installers for Windows and packages for Linux — .exe, .msi, .AppImage, .deb, .rpm. Until now those builds were compiled on every release and discarded. Now they are uploaded and the download page serves the right one.
Add a plugin once and it is reachable from the deck and from the agent CLIs you drive, each written into its own config format. The assistant can call a plugin’s tools directly, and every plugin call asks you first — a tool’s description says what it accepts, never what it does.
Run a task on the weekdays and at the local time you choose, with a history of what ran. Routines are missions, so they inherit the plan gate, the write fence and the spend ceiling. They run while DeckSpace is open — there is no background service, and the app says so rather than quietly missing a slot.
A fresh install registers no hotkeys, opens no microphone, downloads no speech model and runs no listener. One click turns it on, and turning it off actually stops all of it in the running app rather than at the next launch.
First run asks whether you want the strongest model every time, a cheaper one for small edits, or free and local wherever it fits. It defaults to quality. Every pane shows which model is running and the plan card says why that one was chosen.
Set a cap and it is enforced where the work is dispatched, not merely displayed on a chart. Usage is read from the model’s own reported token counts, so what you see is what was billed.
Hold a key to talk, through your desktop’s own shortcuts portal on GNOME, KDE and Hyprland. Your desktop owns the binding, so you set it there. Compositors that ship no portal backend are told so plainly instead of appearing to work.
Agents on Windows run under a restricted token that can only write where you allowed, and the shell guard refuses a destructive git in an agent pane. Both were watched failing a deliberately planted violation on real hardware before this shipped.
Checking on a run from your phone no longer needs the same wifi. The relay is live, enrolment is first-claim-wins, and an install that does not exist is indistinguishable from a Mac that is simply asleep.
Three things we would rather you heard from us. Automatic code review ships switched off. It works, but we know of six open problems with it — the worst being that when two panes finish together one result can be dropped and shown under the other one's title. It stays off until those are fixed; everything above is unaffected. Linux is not downloadable yet. The Linux build works and is part of every release, but we are not handing out installers until it has had more real use. When it does land: no auto-update channel (new versions are a reinstall), X11 rather than Wayland, and no DeckTest browser-testing engine, which is macOS-only. DeckTest on Apple Silicon still needs Rosetta. The app itself is native, but the bundled browser-testing engine is an Intel binary for now, so that one Pro feature asks macOS for Rosetta the first time you use it. Everything else runs native.
Hand a job to the deck and give it a time. It runs in a fresh git worktree, on its own branch, and leaves you a diff to read rather than a change already made. Nightly, on an interval, once at a set time — or right now, if you would rather watch it.
Good for the work that is worth doing and never worth interrupting yourself for: the flaky test that needs bisecting, the dependency bump nobody wants, the audit you keep meaning to run.
Building usually scatters across terminals, tabs and half-finished prompts. DeckSpace gives each part of the job a room — all sharing the same agents, board and memory.
A grid of up to 16 real terminal panes (native PTYs, xterm + WebGL) with layout presets. Launch an agent into any pane and watch its status live. Clickable command blocks with exit-code dots, so you can jump back through your history — in 11 app themes and 37 terminal colour schemes.
Decompose one goal into a team — a git worktree per agent, true isolation — then launch, watch, and merge each result behind a review gate.
Side-by-side diffs, stage & commit, inline comments, and a verification verdict — you decide what ships.
A living markdown knowledge graph with [[wikilinks]], force-directed view and backlinks — shared with every agent over MCP.
Push-to-talk dictation plus an assistant that opens rooms, spawns agents, files board tasks, starts missions, runs verifies, commits, searches code and takes screenshots — and answers out loud.
Hand off a mission and DeckLoop designs, ships and fixes in a loop — then proves the result with a real end-to-end browser test (DeckTest engine, built into Pro).
Plan-gated orchestration where a different provider's model reviews the code before a run counts as done.
A SQLite Kanban that runs its own cards: drag one into In Progress and its agent starts.
File tree, Monaco tabs and quick-open, with optional inline completion using your own key.
Preview your site in a native webview at desktop, tablet and mobile widths, click an element, describe the change, and an agent edits the source.
Search your codebase by meaning when an embedding provider is configured, by keyword when it is not.
Every parallel task runs on its own branch and git worktree, so agents never trip over each other's files. Fill the whole deck, watch it move at once, and merge each result when it's ready — one at a time, on your terms.
The point isn't to hide the work — it's to keep the moving parts visible enough to steer. A verification gate runs tests, lint and a secret scan before anything merges, and every write is confined to the repo.
DeckSpace launches the CLI agents you already have installed and authenticated — mix and match models across panes. No new accounts, no lock-in. Bring your own key; DeckSpace never sees or stores it.
claude
codex
aider
agy
opencode
cursor-agent
droid
copilot
grok
amp
qwen
gemini
DeckSpace is a native desktop app (Tauri + Rust) — small, fast and offline-capable. The terminal, editor, board and memory all run locally. The only network calls are the ones your own agent CLIs make to your own model provider.
Use the CLIs you already pay for. DeckSpace never meters or marks up your model usage — you pay your provider directly. DeckSpace itself is a flat subscription, never per-token.
Confine an agent's writes to the repo, reroute destructive commands to a recoverable trash, and snapshot before every action.
Block cloud-metadata endpoints and SSRF. Telemetry is off by default — nothing about your code goes anywhere.
All three plans start with a 7-day free trial — card required, cancel anytime. Flat fees, bring your own key, $0 metering.
$24/mo, or $19/mo billed annually ($228/yr).
$39/mo, or $31/mo billed annually ($372/yr). Adds DeckVoice, DeckTest and Claude Accounts while you subscribe.
$69 per seat per month, 2 to 50 seats, billed monthly.
Up to 16 terminal panes, each able to host an agent — real native PTYs, not a toy sandbox. Launch one, or fill the deck and watch every pane move at once. Each parallel task gets its own git worktree so they never collide.
Twelve CLI agents today: Claude Code, Codex, Aider, Antigravity, OpenCode, Cursor Agent, Droid, Copilot CLI, Grok CLI, Amp, Qwen Code, and a legacy Gemini entry. You can register any other CLI yourself. DeckSpace detects whichever are on your PATH.
Bring your own key. You authenticate the agent CLIs once, the way you already do — DeckSpace never sees, stores, meters or marks up your usage; you pay your model provider directly. DeckSpace itself is a flat subscription. DeckSpace Pro is $24/mo, or $19/mo billed annually ($228/yr). Deck Complete is $39/mo, or $31/mo billed annually ($372/yr), and adds DeckVoice, DeckTest and Claude Accounts while you subscribe. Deck Complete Team is $69 per seat per month, 2 to 50 seats, billed monthly. All three start with a 7-day free trial (card on file, cancel anytime before day 7 and you're not charged).
No. The terminal, editor, board and memory all run locally and offline. The only outbound calls are the ones your own agent CLIs make to your own model provider. Telemetry is off by default.
Atlas is DeckSpace's voice assistant (Pro). It knows the app: say "open the board", "create a task called fix login", "spawn Claude on the checkout bug", "what's pane 2 doing?", "start a mission to build a landing page and test it" — and it acts, then answers out loud. It can also see your screen and operate other Mac apps, with a spoken confirmation before anything risky.
Yes. Pro bundles the DeckTest engine: missions can finish with a real end-to-end test in a real browser — sign-ups, clicks, assertions. A failing test automatically sends the agent back to fix it.
macOS 12 and newer. It is a native Tauri + Rust app — small, fast and offline-capable. Windows and Linux are not published yet; tell us if you want them and it moves the roadmap.
DeckSpace for macOS is live — start your 7-day free trial and cancel anytime. Bring your own key; no usage markup.