Introduction

The error tensorflowjs_converter: command not found means the shell cannot locate the TensorFlow.js converter executable. In most cases the package is either not installed, installed into a different Python environment, or installed correctly but its scripts directory is missing from PATH.

This is a tooling problem rather than a model-conversion problem. Once you verify the environment and invoke the converter from the same interpreter that installed it, the issue is usually resolved quickly.

What the Converter Is

The TensorFlow.js converter is the command-line tool that transforms supported TensorFlow model formats into artifacts that TensorFlow.js can load in the browser or in Node.js. The executable is commonly exposed as tensorflowjs_converter.

Because that executable is installed through Python packaging, it follows the rules of your active Python environment:

• system Python installs put it in a system scripts directory,
• virtual environments put it inside the virtual environment,
• user installs often place it under a local bin directory that is not always on PATH.

That is why the same command may work in one terminal and fail in another.

Install the Package in the Right Environment

Start by installing the package with the same Python interpreter you plan to use for conversion:

python -m pip install tensorflowjs

After installation, verify that the package is available:

python -m pip show tensorflowjs

If this command does not find the package, the converter was never installed in that interpreter environment. Fix that before checking anything else.

Prefer the Module Form When PATH Is the Problem

If tensorflowjs_converter is not found but the package is installed, run the converter through Python directly:

1python -m tensorflowjs.converters.converter \
2  --input_format=tf_saved_model \
3  --output_format=tfjs_graph_model \
4  /path/to/saved_model \
5  /path/to/tfjs_model

This approach bypasses the shell executable lookup and uses the installed module from the active interpreter. It is often the fastest fix, especially inside virtual environments and CI jobs.

If that command works, your installation is fine and the real issue is just shell path resolution.

Check Virtual Environments and PATH

Many failures happen because the package is installed inside a virtual environment that is not activated in the current shell.

Typical workflow:

1python -m venv .venv
2source .venv/bin/activate
3python -m pip install tensorflowjs
4python -m tensorflowjs.converters.converter --help

On Windows Command Prompt, activation looks different:

.venv\Scripts\activate
python -m pip install tensorflowjs
python -m tensorflowjs.converters.converter --help

If you insist on using the tensorflowjs_converter executable form, make sure the environment's scripts directory is on PATH. On many UNIX-like systems, user installs place console scripts under a directory such as ~/.local/bin.

A Simple Working Example

Suppose you already exported a TensorFlow SavedModel and want a TensorFlow.js graph model:

1python -m tensorflowjs.converters.converter \
2  --input_format=tf_saved_model \
3  --output_format=tfjs_graph_model \
4  ./saved_model \
5  ./web_model

This command should produce files such as model.json and weight shard files under ./web_model. If the converter runs but the conversion fails, that is a separate model-format issue. The original "command not found" error should already be gone at that point.

Common Pitfalls

• Installing tensorflowjs with one Python interpreter and running the converter from another.
• Forgetting to activate the virtual environment before using the command.
• Assuming pip install guarantees the executable is on PATH. It only guarantees the package is installed somewhere.
• Debugging model format flags before proving the converter executable can be launched at all.
• Using sudo pip install casually and creating a messy system-level Python environment.

Summary

• 'command not found means the shell cannot find the converter executable, not that TensorFlow.js conversion is fundamentally broken.'
• Install tensorflowjs with the same Python interpreter you intend to use.
• 'python -m tensorflowjs.converters.converter is the most reliable way to bypass PATH issues.'
• Virtual-environment mismatches are one of the most common causes.
• Once the command launches, any remaining problem is likely about model format or conversion flags, not installation.