TensorBoard in your terminal

Watch training curves from an SSH session.

pip install terminalboard, point it at a logdir, and your live TensorBoard curves render as Unicode/braille text. No browser, no X11, no port forwarding.

PyPI GitHub

terminalboard rendering TensorBoard curves inside a terminal.

Design brief

Checking a curve should feel like tail -f.

terminalboard is a terminal-native TensorBoard viewer with a small AI assistant.

Install

pip install terminalboard

Features

One less browser tab on remote machines.

Reads TensorBoard event files directlyScalars, histograms, text summaries, and PR curvesMulti-run overlays, smoothing, log-Y, zoom, and cursor drill-downA compact filter grammarPress a to chat with your runsWorks with model providers through LiteLLM

Assistant

Press a and chat with your runs.

The assistant sees the live view and the log data, then answers and drives the dashboard in the same turn. Ask for validation losses with smoothing, and the filter is applied. Ask which run is overfitting, and it compares train and validation curves.

The assistant stays off until configured. When enabled, requests send tag names and metric summaries to the provider you choose, and tag names can reveal architecture details. If that matters, point it at a local Ollama model.

Source is on GitHub, and the package is available on PyPI.

Back to products