Objects at a glance — Kaimei API Documentation

Every record carries an object type and a prefixed id. Models and projects also have a human slug that works anywhere an id does.

Object	Id prefix	What it is	Reference
`model`	`mdl_`	A registered model: feature schema, output definition, inference configuration, optional uploaded structure and reference data. Versions are separate models that share a `lineage_id`.	Models
`project`	`prj_`	A folder that groups models and rolls up their traffic and alerts.	Projects
`prediction`	`pred_`	One input and the output a model produced for it, logged by you or made by Kaimei.	Predictions
`explanation`	`exp_`	Feature-level attributions for one input, with the ranked `top_features` and an optional counterfactual.	Explanations
`analysis`	`anl_`	A background study of a whole model. Produces insights.	Analyses
`insight`	`ins_`	One finding about a model: importance, behaviour curves, interactions, robustness, drift, learned rules or output behaviour.	Insights
`organization`	`org_`	Your workspace, its plan and its quotas.	Organization
`api_key`	`key_`	A credential for the organization.	API keys
`user`	`usr_`	A dashboard user. Owners and admins can change organization settings.	Auth
`model.structure`	—	The architecture document of an uploaded model, with its tensors (`tsr_`) and parameters.	Model structure

The usual order of operations:

Register a model (POST /v1/models), or upload a fitted estimator from a notebook, which registers it with its layers and weights.
Give it a population: reference rows (PUT …/reference-data) that explanations compare against and analyses study. Logged predictions work too.
Explain single inputs (POST /v1/explanations), analyse the whole model (POST /v1/analyses) and read the insights.
Log what the model does in production (POST /v1/predictions/batch); run a drift analysis on a schedule; explain any logged prediction by prediction_id.
Ship the next model as a version and compare their insights.