Automations

An automation is a workflow of agents: several agents wired together so each one works on the previous one’s output, in sequence. Instead of one agent doing one job, a chain runs end to end — research feeds a writer, which feeds a publisher — all sharing one memory and governed by your decisions.

​

Two ways to build one

• Just ask Chat. Describe the end-to-end outcome and Orbie builds the chain for you — the fastest path.
• Use the visual canvas. Drag your agents onto a canvas and connect them yourself when you want fine control.

​

The pieces

• Automation — the whole workflow.
• Entry point — the one existing agent that kicks it off.
• Follow-up steps — what happens next: a Claude call, a tool call, a transformation, or handing off to another one of your existing agents.

Only agents already active in your workspace can be placed — automations connect the team you already have.

​

What starts an automation

The entry point can fire on any of:

• A schedule — at a set time or cadence.
• From a previous step — when an earlier step finishes.
• An external event — something happening outside Orbit.

(Any of these can trigger it.)

​

Why this is the unlock

Single agents are useful. Chained agents are a company. A lead-sourcing agent hands enriched leads to an outreach agent; a research agent hands findings to a content agent. Each step builds on the last, nothing is re-derived, and the whole chain reads from the same Vault and obeys the same decisions. This cross-coordination — agents handing off to each other with shared memory — is what turns a pile of separate bots into an operation that runs itself.

​

What can be an entry point

It depends on the agent’s underlying engine:

• Cowork agent — schedule triggers.
• Claude Code routine — all trigger types.
• Codex / others — varies by engine.