Skip to main content
The Antideploy API creates a project, deploys it, and reports on it. It is deliberately small, because that is all deploying takes.

​
Base URL

That URL is also a live, unauthenticated description of the API. curl it and you get the full contract as JSON: request shapes, error codes, and current limits. If you are pointing an AI agent at Antideploy, give it https://antideploy.com/llms.txt and let it find the rest. It needs no credential from you at all.

​
Authentication

Every request carries a bearer token:
There are two kinds. A project key (ad_) is scoped to a single application and identifies it, so no application id is ever sent. An account token (adu_) represents you, can create applications, and therefore has to name the one it is acting on with ?applicationId=<id>.

Authentication

Both credentials, what each can do, rotating and revoking.

Terminal login

How an agent gets an account token without a browser.

​
Endpoints

That is the complete surface. Deleting an application is deliberately not on it: that needs the dashboard, so no credential sitting in a project directory can destroy one.

Deploy over the API

The full deploy reference, with the one command that ships a directory.

​
Request format

POST /api/v1/deploy is multipart/form-data: it carries your source tree. Everything else is JSON.

​
Rate limits

A second deploy while one is running returns 409 with the in-flight taskId, not an error you need to retry blindly. Exceeding the hourly limit returns 429 with retryAfterSeconds.
Both limits exist because agents retry. A human clicks Deploy a few times a day; a loop that treats failure as retryable will hit the API as fast as it is accepted.

​
Errors

Errors return a JSON body with a stable code, a human-readable error, and a documentation URL pointing back at the contract.
Match on code, not on the message text. Two carry more than a message, and are worth reading rather than just relaying: too_large lists every offending path in files, and name_taken returns the applicationId you already have, so a script can use it instead of failing.

​
Asynchronous by design

POST /api/v1/deploy returns 202 immediately with a taskId. Building and releasing take a minute or two. Poll /api/v1/deployments/{taskId} until status is succeeded or failed. The poll response carries warnings and hazards as structured data. A deploy can succeed and still not behave as its author expects; those fields are how you find out.