Using ReOpenly from an agent

An agent, a script or a CI job can run the whole platform with an API key: create projects, load data, train, host, serve, share, search and read billing. Three steps need a person, and a few account settings refuse an API key on purpose. This page lists both, then how to find and call everything else.

What needs a person

Step	Why	What an agent does instead
Create the account	Sign-up happens in a browser and ends with the person accepting the Terms. That acceptance has to be theirs.	Ask the person to sign up at reopenly.com.
Issue the first API key	The first key is created from a signed-in session.	Once it has one key, an agent can create, rotate and revoke keys itself with POST /api/v1/keys.
Pay for credits	POST /api/v1/credits/checkout returns a payment page that a person completes.	Hand the returned url to the person, then poll GET /api/v1/credits/checkout/status.

What an API key cannot change

These answer 403 to an API key, so a leaked key cannot take over the account: your display name, password and email (PATCH /api/v1/me/profile, POST /api/v1/me/password, POST /api/v1/me/email and its verification step), and the privacy choices under /api/v1/me/preferences.

Find the API

• /llms.txt indexes the site, and /docs/llms.txt lists every public operation with a Markdown page each.
• The OpenAPI document, in English and Portuguese. Every operation says what it does, when to use it and what it returns; every JSON request and response carries an example checked against the API; and an operation that needs no key says so with security: [].
• GET /api/v1/me returns your workspace slug, the owner_slug in every project URL you own. GET /api/v1/base-models lists the models you can train and the bounds of every setting.

What an agent can do with a key

Job	Main operations
Use a base model, no project	GET /api/openai/v1/models, then POST /api/openai/v1/chat/completions or POST /api/openai/v1/embeddings (you pay per use; nothing is saved)
Curate chat data	POST .../chats/bulk, GET .../tags
Curate embedding data	POST .../datasets, POST .../examples, file imports
Train	GET .../training-jobs/estimate-messages or POST .../training-jobs/preview, then POST .../training-jobs and GET .../training-jobs/{job_id}
Control a run	cancel before training starts, pause after, then resume or finish
Host and serve	POST .../models, then the OpenAI-compatible routes
Share	POST .../models/{model_slug}/shares; recipients find it in GET /api/v1/models
Search	POST .../search/indexes, then .../query
Collect training data from searches	POST .../retrieval-events, then .../promote
Watch spend	GET /api/v1/credits, GET /api/v1/credits/usage

Retry safely

• Inference: retry automatically only when the error code is model_warming or overloaded. Anything else is permanent, your mistake, or a request that may already have been billed.
• Training: POST .../training-jobs takes no idempotency key. An account runs one unfinished training job at a time, so a second start answers 409 training_slot_occupied; after a timeout, list the jobs before trying again. .../resume and .../retry require an Idempotency-Key header.
• Checkout: send a fresh Idempotency-Key per purchase attempt.
• Imports: .../confirm takes an idempotency_key in the body.
• Recorded searches: the event_id is the idempotency key; replaying the same body returns 200.

Every /api/v1 error is {"detail": {"code", "message"}}: branch on code and show message. See Errors.