Crewdeck: your multi-agent workspace

Your multi-agent workspace

However you code, there's a mode for you.

For Windows 11. Works with Claude Code, Codex, OpenCode, Antigravity, Cline and MiniMax Code, on your own plans and keys.

PLAYER 1 · PRESS START

Give the crew a job

Type something small. The coordinator sizes the job and picks a crew, the board fills up, and the finished app runs right here on the page.

LEVEL 01 / 05

Vibe Code

Up to six coding agents, each on its own project, all on one screen. A kart racer, a haunted house and a website, side by side, and you steer every one.

Panels: 1 to 6, arranged how you like
Each panel: Its own project and conversation
Clients: Claude Code, Codex, OpenCode, Antigravity, Cline, MiniMax Code
Needs you?: The panel says so, and the rest keep going

Real recording, 15 Oct 2026: six agents each building a Kanban board. Sped up 3×.

LEVEL 02 / 05

Collab

One project, one crew. The coordinator sizes the job and proposes a team; nothing starts until you approve it. Then they share one board, one chat and one folder, and the lines show who's waiting on whom.

Crew: Right-sized: one builder for small jobs, up to five for big ones
Planned by: The coordinator, approved by you
Mix: Agents on your plans plus any model you have a key for
Shared: Board, group chat and files

Replay of a real Agent Hub run (test AH-001, 23 Oct 2026), condensed. Point at or tap a card to see what it's waiting on.

The coordinator's proposed team for a recipe app: builder, second builder, UX designer, quality assessor and code reviewer, each with a model and a price note — The coordinator's proposal for a bigger job: five roles, each with a model and what it costs you.

LEVEL 03 / 05

Face Off

One brief, several agents, the same start. Each gets its own copy of the starting folder and a fresh conversation, and the brief goes to all of them at the same moment.

Racers: Every client you tick
Fair start: Same folder, same brief, same second
Records: Every run saved for you to compare

Claude Code, Codex and Antigravity building the same app side by side — Three agents on the same brief, 16 Oct 2026.

LEVEL 04 / 05

Benchmark

Frozen challenges, run one after another while you're away. Every request's tokens, the time, your interventions and the files are recorded. A question, a problem or the time limit pauses the suite and tells you; it never answers for you.

Suite: Seven tests, from a phone OS to a Blender-to-Godot game
Recorded: Timing, tokens, interventions, files
Results: The same runs behind the DF9 Model Gauntlet

A benchmark suite running: Kimi K3.1 Pro in OpenCode on test 1 of 7, with tokens and cost so far — Kimi K3.1 Pro on launch night, 23 Oct 2026: test 1 of 7, every token counted.

LEVEL 05 / 05

Studio

Record any mode for your videos. Put the app being built next to the agent building it, set the text size, hide the controls, and hit recording view.

Shows: Any window beside any agent
Made for: OBS and screen recorders