Agents as code: build and ship Acme agents from your own editor

Most agents start life as a conversation in a chat window. That works well for business teams, but developers spend their day in an editor and a terminal. Meet agents as code: a way to write, run and ship Acme agents without ever leaving the tools you already code in.

Why treat agents like code

Agents that touch production systems deserve the same habits as the rest of your code: version control, review, repeatable tests and a clear path to release. Agents as code keeps the agent definition in a plain file next to your source, so changes show up in pull requests like everything else.

• Define instructions, tools and knowledge sources in a single spec file
• Run the agent locally against real, permission-aware company context
• Publish to your workspace with one command once checks pass

How it works

The Acme CLI reads your spec, validates every tool and data source against your workspace, and starts a local session that behaves exactly like the hosted agent. Test runs record the retrieved documents and tool calls so you can see why the agent answered the way it did.

acme agents init support-triage
acme agents run support-triage --input "Summarize open P1 tickets"
acme agents publish support-triage

Publishing creates a new agent version and keeps the previous one available for rollback. Admins see the same review and approval flow they use for agents built in the app.

Get started

Agents as code is available today for all workspaces on the Enterprise plan. Read the developer guide to set up the CLI, or talk to your account team about early access to the editor extensions.