Developers
Five steps, not thirty.
Create a workspace, create an agent, add a tool, run it in the sandbox, deploy it. No cluster to provision, no queue to configure, no infrastructure decisions before you have written anything.
The sandbox is the default
A new agent starts in a sandbox with synthetic customers and mock tools. You cannot accidentally email a real person on your first attempt, because there is no route from a sandbox to production.
One tool is enough to be useful
The quickstart agent reads an invoice and writes a summary. Real, permissioned, audited — and small enough that when it works you know why.
Promotion is explicit
Moving to production is a deliberate step with the review screen. There is no one-click path from a laptop to a customer's mailbox.
Nothing here is thrown away
The agent you build in the quickstart is a real agent with a real manifest. It becomes the first thing in your repository rather than a tutorial you delete.
Roughly what it looks like
Illustrative and deliberately short. The reference is at docs.agentlabora.com, and pretending a marketing page is documentation wastes an afternoon.
The whole quickstart
bashlabora init invoice-summary
labora tools add invoice.get
labora test --sandbox
labora deploy --env stagingWhere it stops
Fifteen minutes gets you a deployed agent doing something small. It does not get you a production accounts-receivable process — that needs permissions decided, approval thresholds set and an owner named, which is a conversation with your finance team rather than a command.
Developer platform
The rest of it.
Give AI a job.
Build an agent, choose what it can access, define what it may do, and deploy it to work safely across your business.
No code required · Full SDKs and API · Stop any agent instantly · [email protected]