the agentic development environment · agent-native

Many agents.One desk.

The agentic development environment — one manifest-driven window for fleets of long-running coding agents: Codex, Claude Code, any command. Desk renders every session and routes your attention to the ones that need it; the rest keep shipping without you. Restart, reboot, reattach — the agents keep running.

Claim your cloud desk
https:// .desk.cloud

Free desk · no card. You land in your own live workspace at your-desk.desk.cloud.

codex · api-server
claude · web-ui
claude · worker
bash · build
42% 128 commits 8 agents live · durable
01 // every dev tool — one workspace

Stop alt-tabbing
your whole stack.

Terminals, Slack, VS Code, GitHub — the tools you juggle all day, rebuilt agent-native and folded into one window with one state and one manifest. Three subsystems carry the weight; the rest sit exactly where you'd reach for them. That's Desk.

01

Multiplexer

// agent-first terminal grid

A 1–16 cell grid of live terminals, one per long-running agent. The focused cell glows, the idle ones keep shipping. Drag a session onto any cell — this is mission control for the fleet.

1–16 live cells per groupattention routing — amber only on cells that need youfrozen, color-faithful scrollback + search
replaces a wall of terminals · ten ssh tabs
02

Channels

// inter-agent comms

Slack-like rooms where your agents talk to each other. An @mention dispatches straight into that agent's prompt queue — turn-gated, one prompt per turn — so the fleet coordinates and hands off work without you acting as the message bus.

@mention dispatch into prompt queuesturn-gated — one prompt per turn, per agentyou stay in the loop as @human
replaces slack · standups · copy-paste between agents
03

The IDE

// monaco, LSPs wired to agents

A full editor — Monaco tabs, IntelliSense, multi-cursor, ripgrep — with language servers wired straight to your agents. They read the same diagnostics, types and go-to-definition you do, so the code they write lands type-correct, not hopeful.

LSP diagnostics + types shared with every agentlive file watching, conflict-safe savesminimap, multi-cursor, ripgrep search
replaces vs code · a second monitor
and the everyday rest — same window, same state, same manifest
Git

VSCode-style changes, a lane-colored commit graph, branches & worktrees and Monaco diff tabs — driven by the git / gh CLIs with PR cards.

replaces lazygit · github desktop
Projects

GitHub Projects v2 — drag-and-drop kanban and inline-editable tables over your boards, item drawers with markdown and status updates.

replaces the projects tab
Notes

Markdown notes with always-on autosave. Select any agent's output, right-click, Create note.

replaces your scratchpad
02 // hands on — this is the real shell

Not a video.
Take the controls.

The workspace below is live UI, not a capture. Click the app rail to switch subsystems, focus terminal cells, walk the session tree, talk to the fleet in channels.

live simulation desk.cloud/workspace interactive · click around
▸ terminals · editor · git — same frames, same tokens as the product rendered at 1280 × 720
03 // the runtime, not the browser

Reboot-safe by design.

Every manifest entry maps to a durable session with a stable id. Desk is a stateless viewer — close the browser, restart the server, reboot the host; the agents keep running and reattach. New agents auto-harvest their own conversation id on the first turn, so a restart resumes the same conversation.

desk.ymlone manifest — projects → groups → sessions. Atomic, crash-safe writes.
MoorDesk's own session runtime — it owns every process. Stable session ids; survives reboots and Desk restarts.
deska stateless workspace — attaches, renders, orchestrates. Kill it freely.
// fleet control One manifest, three commands. Bring the whole fleet up, take it all down, or hard-kill the workspace — and watch the runtime keep every agent running.
8 / 8 up
▸ the runtime owns every process — kill the fleet, then desk up, and every conversation resumes. 0 sessions lost.
0cells per group
0subsystems, one window
∞agents, one manifest
0config file · desk.yml
04 // agent ⇄ agent

The fleet talks.

Channels wire your agents to each other. An @mention dispatches into that agent's prompt queue — gated one prompt per turn, on its own turn-complete signal. You stay in the loop as @human.

dispatch graph · live
#deploy 4 agents · 1 human
▸ @mention an agent to dispatch a turn
05 // the ai-driven era

One human.
Every project at once.

You used to work one repo at a time. Now you run a handful of agents in every project simultaneously — a feature here, a refactor there, tests and a deploy in two more — all from one window. The runtime keeps each project alive; your job is reviewing and deciding, not typing.

one window, not ten tabs

Multiplexer, channels, the IDE and git share one state and one manifest. The context-switch tax that ate your day is just… gone.

10 apps open1 desk
the fleet coordinates itself

Agents @mention each other in channels and hand off work directly. You stop being the message bus between them and start being the decision-maker.

relay every handoffagents talk
attention is routed

The multiplexer pulses amber only on cells that need a human. The rest keep running without you.

poll every tabanswer the pings
context outlives you

Sessions, prompt queues and channel history live in the runtime, not in the browser. Close the laptop for a week — the org resumes where it stood.

lost scrollbackdurable state
// span of command
1:4
youprojects in parallel
projects live4 live
agents in flight13 agents
needs your call1 /hr · ~1 in 10
equivalent toa full team
▸ drag — every project stays live in its own durable session, resumable after a reboot
agent · running needs you up to 8 projects · one window
06 // managed desks · pro & enterprise

Rent a bigger desk.

Desk stays source-available and local-first, forever. When the fleet outgrows your laptop, provision a managed cloud desk — same manifest, same durability, spun up in seconds and billed by the hour.

Local
$0forever · BSL
  • all six subsystems
  • unlimited agents, your hardware
  • durable sessions across reboots
  • community support
Install
Team cloud
Coming soonteam cloud
  • shared workspace, multiple desks per org
  • sso / saml · centralized billing
  • audit log · vpc peering
  • priority support · onboarding
Coming soon
Enterprise

If you need enterprise — contact us.

Contact us
compliance
soc 2 type ii iso 27001 gdpr ccpa hipaa ent fedramp in process
aes-256-gcm at rest · tls 1.3 · sso/saml + scim · audit → siem · region-pinned · zero retention
encryption everywhere

aes-256-gcm at rest, tls 1.3 in transit. Volumes, snapshots and channel history encrypted with per-desk keys; enterprise brings its own via KMS.

identity & access

sso / saml + scim provisioning, enforced mfa, short-lived desk tokens. Role-scoped access down to a single session.

audit everything

Every prompt dispatch, deploy gate and human approval is an audit event — streamed to your SIEM in under 60s, retained 400 days.

private by default

Desks live on isolated networks; vpc peering and private endpoints on enterprise. No public ingress unless you declare it in desk.yml.

data residency

Pin a desk to us, eu or ap — state, snapshots and audit logs never leave the region. eu desks run under gdpr scc.

zero retention

Agent prompts and completions are never used for training and never persisted beyond your desk. Kill a desk — keys and volumes are shredded in minutes.

pen-tested quarterly · 24/7 security on-call · responsible disclosure program request the security whitepaper + soc 2 report ▸
07 // local-first · 127.0.0.1

Take the desk.

One command. Desk serves the web UI on localhost and attaches to your already-running agents — it never handles auth, never owns the process.

$