TensorBoard
tensorflow.org
- Free plan
- Windows
- macOS
- Linux
- Web
- API
- Self-hosted
What TensorBoard is
TensorBoard is a suite of visualization tools for understanding, debugging, and optimizing TensorFlow programs during machine learning experiments. It tracks metrics such as loss and accuracy, and its Graph Explorer lets users inspect TensorFlow model graphs. Dashboards can show how tensor distributions change over time using histograms and percentile summaries. TensorBoard also displays logged images, audio, and text, projects high-dimensional embeddings, and includes profiling tools for TensorFlow programs. In Keras workflows, a TensorFlow callback can record metrics, training graph visualizations, weight histograms, and sampled profiling data. Users can launch it from the command line or in notebooks with the `%tensorboard` line magic. The project is designed for offline use, including locally, behind a corporate firewall, or in a datacenter. TensorBoard 2.0 and later binds to localhost by default; `--bind_all` makes it available on public networks over IPv4 and IPv6. Chrome and Firefox are named as supported browsers. TensorBoard is free, self-hostable, and licensed under Apache-2.0.
Who it suits
It suits people working with TensorFlow experiments who need to inspect metrics, model graphs, distributions, or profiling data. Its offline and self-hosted options also suit teams running it locally or within a datacenter.
Good to know, and what to watch
- Visualizes loss, accuracy, and other metrics.
- Graph Explorer displays TensorFlow model graphs.
- Can display logged images, audio, and text.
- Runs offline and can be used in a datacenter.
- Free, with Apache-2.0 licensing.
- Google Cloud Storage log directories need a TensorFlow installation.
- Other browsers may have bugs or performance issues.
- `--bind_all` exposes it on public networks.
Plans
Every plan and priceTensorBoard
Free plan
The verdict
TensorBoard brings experiment visualization and profiling into a tool designed for local and offline use. Users should note its browser guidance and take care when changing the default localhost binding.
What the maker publishes
- Browser support
- The README says TensorBoard can be used in Google Chrome or Firefox and that other browsers may have bugs or performance issues. source
- Histograms and distributions
- Dashboards show how tensor distributions change over time, including histograms and percentile summaries. source
- Keras integration
- The TensorFlow Keras callback records metrics, training graph visualizations, weight histograms, and sampled profiling data. source
- License
- The TensorBoard GitHub repository identifies its license as Apache-2.0. source
- Metrics
- It tracks and visualizes metrics such as loss and accuracy. source
- Model graphs
- Its Graph Explorer visualizes TensorBoard graphs to inspect TensorFlow models. source
- Notebook integration
- TensorBoard can be launched in notebooks using the `%tensorboard` line magic or from the command line. source
- Offline and hosting
- The project README says TensorBoard is designed to run offline and can be used locally, behind a corporate firewall, or in a datacenter. source
- Other data
- TensorBoard can display logged images, audio, and text, and project high-dimensional embeddings. source
- Profiling
- TensorBoard includes tools for profiling TensorFlow programs. source
- Purpose
- TensorBoard is a suite of visualization tools for understanding, debugging, and optimizing TensorFlow programs during machine learning experimentation. source
- Security setting
- TensorBoard 2.0 and later defaults to `--host localhost`; `--bind_all` makes it serve on the public network over IPv4 and IPv6. source
- Storage limit
- The README says TensorBoard does not support log directories on Google Cloud Storage when run without a TensorFlow installation. source
- Support
- The TensorBoard README directs general usage questions to Stack Overflow and bug reports to GitHub Issues. source
Checked against the maker's pages on 30 Sept 2026.