API

DashAI exposes a RESTful API at /api/v1. All endpoints return JSON. The API is built with FastAPI and supports OpenAPI documentation at /docs (Swagger UI) and /redoc.

Router Structure

Router file	Resource	Purpose
components.py	/api/v1/component	List and filter registered components
datasets.py	/api/v1/dataset	Upload, list, and validate datasets
model_sessions.py	/api/v1/model-session	CRUD for model training sessions
runs.py	/api/v1/run	CRUD for training runs, metrics, optimization plots
jobs.py	/api/v1/job	Enqueue jobs and query job status
explorers.py	/api/v1/explorer	Launch and retrieve data explorations
explainers.py	/api/v1/explainer	Launch model explanations
converters.py	/api/v1/converter	Apply data transformations
predict.py	/api/v1/predict	Run predictions on new data
plugins.py	/api/v1/plugin	Manage installed plugins
generative_session.py	/api/v1/generative-session	Generative model sessions
generative_process.py	/api/v1/generative-process	Generative process execution and results

Key Endpoints

Components

GET /api/v1/component/
GET /api/v1/component/?select_types=["Model","Metric"]

Returns all registered components with their schemas and metadata. Accepts an optional select_types query parameter to filter by component category. The frontend uses this response to render configuration forms dynamically.

Datasets

GET    /api/v1/dataset/
POST   /api/v1/dataset/
GET    /api/v1/dataset/{dataset_id}
PATCH  /api/v1/dataset/{dataset_id}
DELETE /api/v1/dataset/{dataset_id}
GET    /api/v1/dataset/{dataset_id}/sample
GET    /api/v1/dataset/{dataset_id}/info
GET    /api/v1/dataset/{dataset_id}/types
GET    /api/v1/dataset/{dataset_id}/model-sessions-exist
GET    /api/v1/dataset/filter/
POST   /api/v1/dataset/copy
PATCH  /api/v1/dataset/{dataset_id}/columns/rename
GET    /api/v1/dataset/export/csv

POST creates a dataset entry (upload is handled as multipart form data). PATCH renames a dataset. The /sample endpoint returns 10 representative rows; /info returns schema and column metadata; /types returns per-column data types. /filter/ supports pagination and column filtering for the frontend data grid. POST /copy duplicates an existing dataset. /export/csv downloads the full dataset as a CSV file.

Model Sessions (Experiments)

GET    /api/v1/model-session/
POST   /api/v1/model-session/
GET    /api/v1/model-session/{session_id}
PATCH  /api/v1/model-session/{session_id}
DELETE /api/v1/model-session/{session_id}
POST   /api/v1/model-session/validation

A ModelSession captures the full experiment configuration — dataset, task, input/output columns, split ratios, and selected metrics. POST /validation checks that the selected dataset columns are compatible with the chosen task before the session is created.

Runs

GET    /api/v1/run/
POST   /api/v1/run/
GET    /api/v1/run/{run_id}
PATCH  /api/v1/run/{run_id}
PATCH  /api/v1/run/{run_id}/reset
DELETE /api/v1/run/{run_id}
DELETE /api/v1/run/{run_id}/operations
GET    /api/v1/run/{run_id}/operations/count
GET    /api/v1/run/plot/{run_id}/{plot_type}

Each Run belongs to a ModelSession and holds the model name, parameters, optimizer config, and training artifacts. PATCH /{run_id}/reset resets a run back to NOT_STARTED. /operations/count returns the number of explainers and predictions associated with a run; DELETE /operations removes them all. /plot/{run_id}/{plot_type} fetches hyperparameter optimization plots — plot_type is one of history, slice, contour, or importance.

Jobs

POST   /api/v1/job/
GET    /api/v1/job/status/{job_id}

All long-running operations (training, exploration, prediction, conversion) are submitted as jobs. POST enqueues a job and returns a job ID immediately. The frontend polls GET /status/{job_id} until the job reaches finished or error.

Explorers

GET    /api/v1/explorer/
POST   /api/v1/explorer/
GET    /api/v1/explorer/{explorer_id}/
PATCH  /api/v1/explorer/{explorer_id}/
DELETE /api/v1/explorer/{explorer_id}/
POST   /api/v1/explorer/{explorer_id}/results/
PUT    /api/v1/explorer/{explorer_id}/results/

Explorers run visualizations (scatter plots, histograms, etc.) on a Notebook's dataset. POST / creates an Explorer record and enqueues an ExplorerJob. POST /{explorer_id}/results/ retrieves the computed result; PUT updates it. Results are stored as Plotly JSON specifications.

Converters

POST   /api/v1/converter/
GET    /api/v1/converter/{converter_list_id}
GET    /api/v1/converter/notebook/{notebook_id}
DELETE /api/v1/converter/{converter_list_id}

POST saves a single converter step to a Notebook. GET /notebook/{notebook_id} returns all finished converter records for that Notebook. DELETE reverts the Notebook dataset to the state before the deleted converter and re-enqueues all preceding converters.

Explainers

GET    /api/v1/explainer/global
POST   /api/v1/explainer/global
GET    /api/v1/explainer/global/{explainer_id}
GET    /api/v1/explainer/global/plot/{explainer_id}
DELETE /api/v1/explainer/global/{explainer_id}

GET    /api/v1/explainer/local
POST   /api/v1/explainer/local
GET    /api/v1/explainer/local/{explainer_id}
GET    /api/v1/explainer/local/plot/{explainer_id}
DELETE /api/v1/explainer/local/{explainer_id}
POST   /api/v1/explainer/local/validate-dataset

PATCH  /api/v1/explainer/

Global explainers (e.g., Permutation Feature Importance) produce a single explanation covering the whole model. Local explainers (e.g., KernelShap) produce per-instance explanations. Both support a /plot endpoint that returns the visualization. POST /local/validate-dataset checks that an input dataset is compatible with the trained run before creating a local explainer.

Predictions

GET    /api/v1/predict/
POST   /api/v1/predict/
DELETE /api/v1/predict/{prediction_id}
GET    /api/v1/predict/filter_datasets
POST   /api/v1/predict/preview

POST / creates a persisted prediction job linked to a trained Run. GET /filter_datasets returns only those datasets whose column schema is compatible with the run's training dataset. POST /preview runs a synchronous prediction and returns the result immediately without persisting it — useful for quick interactive inference.

Generative Sessions and Processes

GET    /api/v1/generative-session/
POST   /api/v1/generative-session/
GET    /api/v1/generative-session/{session_id}
PATCH  /api/v1/generative-session/{session_id}
DELETE /api/v1/generative-session/{session_id}

POST   /api/v1/generative-process/
GET    /api/v1/generative-process/{process_id}
GET    /api/v1/generative-process/{process_id}/results

A GenerativeSession stores the model and parameters for a generative workflow (text-to-text, text-to-image, ControlNet, etc.). PATCH updates the session parameters and records a snapshot in the parameter history. Each invocation creates a GenerativeProcess record that tracks status and stores input/output payloads via ProcessData records. GET /{process_id}/results returns the output after the process completes.

Dependency Injection

Endpoints receive dependencies through FastAPI's Depends mechanism combined with Kink's di container:

@router.post("/")
@inject
async def enqueue_job(
    job_params: JobParams,
    session_factory: sessionmaker = Depends(lambda: di["session_factory"]),
    component_registry: ComponentRegistry = Depends(lambda: di["component_registry"]),
    job_queue: BaseJobQueue = Depends(lambda: di["job_queue"]),
):
    ...

Internationalization

The API supports multilingual responses. Component display names and descriptions are stored as MultilingualString objects, and the API filters them based on the Accept-Language header.

Interactive Docs

When running DashAI locally, you can explore the full API interactively:

• Swagger UI: http://localhost:8000/docs
• ReDoc: http://localhost:8000/redoc