Multi-Node Relay

Multi-node connects every machine signed into the same Sivants account, so any of them can queue prompts and run tools on the others — a Windows PC and a Mac can be driven from the same interface, with each running the prompts it’s suited for.

Machines as peers

Each machine is a first-class node on your account. From Settings → Account → Connected Machines, a “Run on: [machine ▾]” picker lets you queue a prompt on one machine that targets another; the prompt executes remotely and streams its output back to you. If a connection drops mid-run, Orbit reconciles on reconnect using a ring buffer and a prompt:sync step, so you don’t lose the output. Each project can carry a default execution target.

Primary node and project authority

One machine is marked primary (Settings → Account → Connected Machines → Set as primary); node-local MCP tool calls route there by default. Project authority gives each project one owning machine, prompt numbering is shared across every machine on the account, and each run gets its own isolated worktree. Keeping each project on one machine is the simplest setup, but working on the same project from several machines is supported as long as git stays clean and pushed.

Two-tier MCP routing

One account-wide MCP URL routes tool calls to any machine through the relay. Account-scoped tools need no node parameter; node-local tools default to your primary node but accept an optional node parameter to target a specific machine. Per-machine MCP URLs remain available as a fallback, and you manage the tokens under Connected Machines → MCP Tokens.

Bridge

Bridge is a related idea at the instance level: a human-gated message channel between two AI instances, where every drafted reply waits in an Outbox until you Approve & send and participants are read-only. It ships as a visual preview on mock data, with the live relay wired in a follow-up.

Next

• Set Up Multi-Node
• Multi-Node, Relay & Bridge reference