REST API Overview

miStudio's backend exposes a full REST API — everything the UI does, you can script. The interactive Swagger UI at http://<backend>:8000/docs is generated from the same code and is always current; these pages add organization and context.

Base URL

All endpoints are rooted at:

/api/v1

Through the standard nginx/ingress deployment this is same-origin with the frontend (e.g., http://mistudio.example.com/api/v1/datasets); hitting the backend directly, it's port 8000.

Conventions

• Async job pattern: endpoints that start heavy work return 202 Accepted immediately with the created record; progress arrives via WebSocket or by polling the record's GET endpoint. The record carries status, progress (0–100), and error_message.
• Errors return structured JSON with an appropriate status code and a detail message.
• Pagination: list endpoints accept ?page=1&limit=50 (or skip/limit) and return totals in the response body.

Common status codes

Code	Meaning in miStudio
200	Success
201	Resource created
202	Background job accepted — track via WebSocket or polling
204	Deleted / no content
404	Resource not found
409	Conflict — e.g., deleting a model that a training still references
410	Endpoint removed — you're calling a deprecated path (hard-refresh the frontend)
422	Validation error (FastAPI/Pydantic detail included)
503	Dependent service unavailable — e.g., labeling LLM has no model loaded

Endpoint groups

Group	Prefix	Page
Datasets & tokenization	/datasets	Datasets
Models & activation extraction	/models	Models
SAE training	/trainings	Trainings
SAE management	/saes	SAEs
Features & labeling	(no prefix) /features, /extractions, /labeling	Features & Labeling
Steering	/steering	Steering
Circuits & clusters	/circuits, /circuit-capture, /circuit-discovery, /validation-manifests, /cluster-profiles	Circuits & Clusters
Neuronpedia export & push	/neuronpedia	Neuronpedia
System monitoring & task queue	/system, /task-queue, /workers	System
Templates & settings	/*-templates, /settings	Templates & Settings