Skip to content

Documentation

Wizard, the Dydema EEG API

Describe an EEG classification task, send labeled trials, and get back a hosted classifier chosen by how well it does on your own data. One key, one base URL, one error shape, and a usage receipt on every response.

Build with an AI assistant

Help me use the Dydema EEG API in this project: developers.dydema.com/llms.txt

What it is

The API is Wizard, a mixture model for EEG classification: licensed pretrained models and classical methods, and a procedure that picks among them for each task. You describe the task and send labeled trials. Wizard selects the methods that fit your montage and sampling rate; each is fitted and scored on your own trials, and the best is hosted as your task model. New trials sent to its prediction route come back with a label and the probability of every class.

Wizard accepts any sampling rate, any trial length and any 10-10 or 10-20 layout, so you can send what your amplifier recorded. It supports event-related potentials, motor imagery and movement, and sleep staging, and fits other tasks on your data; it refuses emotion recognition and clinical uses. Every response says what it cost.

Base URLhttps://dydema--eegapi-gateway.modal.run/v1
AuthenticationAuthorization: Bearer $DYDEMA_API_KEY (details)
Task models/v1/task-models: create, predict, list, expire, delete (guide)
UsageGET /v1/usage (details)
FormatJSON request and response bodies
Errors{type, code, message, field, alternatives, details} (every code)
RegionUnited States

What happens to a request

  1. Authentication. Your key is checked; a revoked key stops working on the next request.
  2. Limits. Permission, request size, rate limit, spending cap and concurrency are checked in that order; each refusal has its own code and costs nothing.
  3. Selection. On a create, Wizard reads your description and the dataset’s shape (channel names, rate, trial length, classes), never the signal, and chooses the methods to test.
  4. Test. Each method is fitted on part of your trials and scored on the rest; the best is refitted on all of them and hosted (how trials are prepared).
  5. Receipt. The response carries the result (a task model with its training_report, or one label per trial) and what it cost, with a request id in X-Request-Id.
A task model costs a one-time create fee, which includes the fit, and flat monthly hosting; predictions are billed per window ($0.003) (details).

Where to go

QuickstartKey to your first task model and its predictions.AuthenticationAPI keys, permissions, and replacing a key.Use with AI assistantsOne page to paste into Claude or ChatGPT.WizardThe mixture model: how it picks an encoder for your task, what it takes, what it returns, and its limits.Dataset formatTrials, labels and channel names for a task model, with EDF, BIDS and NumPy conversion scripts.What Wizard draws fromAutomatic method selection, compatibility and evaluation.Signals & montagesAny sampling rate, trial length and 10-10 or 10-20 layout, and what each kind of encoder does with it.Errors & limitsEvery error code, rate-limit headers, retrying.Usage & billingThe create fee, hosting, predictions, and how to read your usage.API referenceEvery route and field, generated from the API's OpenAPI document.