Back to all features local-network collaboration

Your whole team, one live canvas.

Everyone edits the same DISPATCH canvas in real time over your LAN — named cursors, inline comments, and live node-wiring. Local‑first: no cloud, no account, nothing leaves your network.

Active named cursors ·inline comments ·live wiring ·no cloud
live demo

Four teammates. One graph. Zero refresh.

A live render, not a screenshot. Cursors roam the shared canvas, kai is wiring Plan → Create, and a comment is pinned to a node. The feed on the right is the presence stream every peer receives over the LAN.

shared canvas · lan 4 connected
ana kai sam you 192.168.1.0/24
Investigate web_search
Plan opus
Create draft
Review gate
ana
kai
sam
you
peers: ana · kai · sam · you · CRDT merge · last-write-wins on style · syncing
presence stream /canvas/ops
kaiwired Plan → Createnow
kaicommented on Plan2s
anamoved Investigate4s
samset Create model · draft7s
youselected Review9s
anajoined · 192.168.1.4114s

Every action is a CRDT op multicast to peers on the subnet — no central server arbitrates.

what it is

Real-time multiplayer, kept on your network.

Open the same workspace from any machine on your LAN and you edit the exact same graph as everyone else. Edits, wiring, model swaps, and comments land for all peers within a frame or two.

named cursors

See who's where.

Each peer carries a colour and a name tag. Watch ana hover Investigate while sam drags Create into place — presence is visible, never guessed.

ana kai sam
inline comments

Pin a thread to any node.

Drop a comment on a Plan, Review, or Merge node and it travels with the graph. Reviews happen in place — no pasting node IDs into a chat elsewhere.

kai gate this before Deploy?
live wiring

Build the chain together.

When a teammate draws an edge — Parallel fanning out to Design and Doc — it streams onto your screen as they pull it. Two people can wire opposite ends of one graph at once.

how it works

Peer-to-peer, over the subnet.

No relay, no broker, no SaaS backend. One peer hosts; the rest discover and join over the local network. Edits flow as conflict-free ops, so the graph converges identically everywhere.

01 · discover Find peers on the LAN

A host advertises the session over mDNS / local discovery. Teammates on the same subnet see it and join — no links to paste, no logins.

02 · sync Snapshot the canvas

On join you receive the current graph state — every node (Initialiser → … → Deploy), its model, edges, and pinned comments — in one transfer.

03 · stream Multicast every op

Cursor moves, edge-wiring, model swaps and comments become small CRDT ops, broadcast to peers and applied as they arrive.

04 · converge Same graph, everywhere

Conflict-free merge means concurrent edits never clobber each other. Drop the network and you keep editing — ops reconcile on reconnect.

session
transportLAN multicast · peer-to-peer
discoverymDNS / local
mergeCRDT · conflict-free
presencenamed cursors + comments
cloudnone · nothing leaves the subnet
why it matters

Designing an agent workflow is a team sport — someone shapes the Plan, someone wires Investigate → Create → Review, someone gates the Deploy. One live canvas beats screen-shares and exported diagrams.

And because it is local-first, it works in an air-gapped lab, on a plane, or behind the strictest firewall — your prompts, graph, and runs never round-trip to the cloud.

How local-first security works
open source · MIT local-first

Bring the
whole team in.

Clone DISPATCH, open a canvas, and let everyone on your network join the same graph. No accounts, no cloud, no cost.

backed by
Pyca Ireland Helios Labs