TENSORVIZ / EARLY PREVIEW
Open your
own model.
Install the extension in local VS Code.
Explore your PyTorch
architecture in the browser.
YOUR FIRST GRAPH
From a Python file
to a clearer view.
-
Install the preview
Download the VSIX above. In VS Code, open the Command Palette and choose Extensions: Install from VSIX…. Select the downloaded file.
-
Open a local model folder
Open your PyTorch project. To try the same model as the website, save the public example source in a new folder and open it in VS Code.
-
Open the class as a graph
Choose Open as Graph above your
nn.Module, select it in TensorViz’s Models view, or run TensorViz: Open as Graph. -
Choose your Python environment
Select an existing Conda or virtual environment. TensorViz checks compatibility and asks before installing missing libraries. Existing Torch/CUDA and other installed package versions are preserved.
If needed, choose Use a separate TensorViz environment. That option downloads a managed environment after consent.
-
Explore first. Validate when you choose.
The graph opens in your default browser. Select layers, expand a module, and reveal its source. Opening a graph does not execute your model.
To inspect runtime shapes, save your model and dependencies, then choose Validate. A successful result applies to that input and source version.
REVIEWING A CHANGE
Your Python stays yours.
Select a layer and choose Edit source…, then Review in VS Code. Inspect the complete Python diff and choose Apply source change or Discard proposal in the native toolbar.
Apply changes the unsaved buffer. Native Undo restores it. Save explicitly before validation; source review checks syntax and graph consistency, not runtime tensor correctness.
CONNECT YOUR CODING AGENT
The model, in your agent’s context.
Use an external client that supports MCP Streamable HTTP and a configurable Authorization header. TensorViz does not include a chatbot or require a provider key.
- Open a model in TensorViz.
- Run TensorViz: Connect coding agent (MCP) from VS Code’s Command Palette.
- Copy the temporary endpoint URL and Authorization header into your client’s HTTP MCP configuration.
- Ask the agent to list the open models and inspect the selected layer. Proposals still go through native Apply/Discard review; validation remains an explicit action.
Keep the connection details private. TensorViz: Disconnect coding agents (MCP) revokes the session. Reconnect after closing or reloading VS Code.
SUPPORT & CURRENT LIMITS
A few things to know.
The graph will not open
Use TensorViz: Select Python Environment and choose
a compatible Python 3.11+ environment. Check the TensorViz output
channel for the reported dependency or parsing issue. Leave
developer settings tensorviz.repoPath and
tensorviz.devServerUrl empty for this packaged preview.
Validation is refused or shows stale source
Save the model and its Python dependencies first. Validation requires a trusted workspace and the selected environment’s PyTorch and model dependencies. Fix Python syntax errors in VS Code before proposing a graph edit.
Which editors and platforms are supported?
The current target is local VS Code. Installed workflow evidence exists for macOS Apple Silicon. Windows x64 and Linux x64 still need certification. Intel macOS has no managed-runtime asset. Remote SSH, WSL and dev containers are unsupported; other compatible editors are unverified.
What is included in a shared graph?
A .tensorviz.json snapshot preserves architecture,
layout and saved explanations. It excludes model source and weights.
File sharing has no account, automatic synchronization or public
model-publishing service.
PACKAGE DETAILS
This is the existing TensorViz QA preview package prepared on September 14, 2026. It bundles the extension, local webpage and backend. No source checkout or Node.js installation is required.