API Reference

API Reference

Overview

TokenPrint uses a clean, documented JSON API for all communication between the frontend and the PyTorch backend. The backend is completely headless and can be used independently of the UI.

Why it matters

If you want to build your own visualizer, write a script to automate tensor analysis, or integrate TokenPrint's data pipeline into a Jupyter Notebook, you can query these APIs directly.

How TokenPrint implements it

The API is served via FastAPI on http://localhost:8000.

• REST API: Synchronous endpoints for fetching static data like model architecture, health status, and full forward-pass traces.
• WebSocket Events: The asynchronous streaming protocol used for Live Inference.
• Data Models: The strict JSON schemas (defined by Pydantic) ensuring consistent types.
• Tensor Format: How multidimensional PyTorch tensors are serialized into flat JSON arrays for web consumption.

Diagram

flowchart LR
    Client([HTTP Client / Script]) --> API[FastAPI: localhost:8000]
    
    API -- GET /architecture --> REST[JSON: Tensor List]
    API -- WS /ws/generate --> WS[Stream: Token Frames]
    API -- POST /analyze --> POST[JSON: Logit Lens & PCA]

Loading

Related pages

• Architecture
• Developer Guide

Further reading

• API Documentation

Navigation

Previous	Home	Next
Performance Tips	Home	REST API