PyGWalker Computation Modes

Use computation to choose where PyGWalker runs data queries. kernel_computation, cloud_computation, and use_kernel_calc are legacy compatibility flags and are scheduled for removal in PyGWalker 0.7.0.

import pygwalker as pyg
 
walker = pyg.walk(df, spec_path="./gw_config.json", computation="kernel")

Mode matrix

`computation` value	Internal behavior	Use when
omitted	Preserve automatic behavior	You want PyGWalker to keep its default selection.
`"auto"`	Preserve automatic behavior	You want to be explicit without forcing a backend.
`"browser"`	Frontend/browser computation	You need static HTML or have smaller local data.
`"kernel"`	Local DuckDB-backed Python computation	You need larger local data in a live notebook, Streamlit, Gradio, or webserver session.
`"cloud"`	Kanaries cloud computation	You want cloud-backed computation with a Kanaries API key.

Recommended examples

For static or frontend-only output, choose browser computation.

html = pyg.to_html(df, spec_path="./gw_config.json", computation="browser")

For a live notebook or app session with larger local data, choose kernel computation.

walker = pyg.walk(df, spec_path="./gw_config.json", computation="kernel")

For cloud computation, choose cloud mode and pass your API key.

walker = pyg.walk(
    df,
    spec_path="./gw_config.json",
    computation="cloud",
    kanaries_api_key="...",
)

Legacy flags

Legacy option	Replacement	Status
`kernel_computation=True`	`computation="kernel"`	Deprecated, scheduled for removal in PyGWalker `0.7.0`.
`kernel_computation=False`	`computation="browser"`	Deprecated, scheduled for removal in PyGWalker `0.7.0`.
`cloud_computation=True`	`computation="cloud"`	Deprecated, scheduled for removal in PyGWalker `0.7.0`.
`use_kernel_calc=True`	`computation="kernel"`	Deprecated, scheduled for removal in PyGWalker `0.7.0`.
`use_kernel_calc=False`	`computation="browser"`	Deprecated, scheduled for removal in PyGWalker `0.7.0`.

Legacy flags still work for compatibility and emit deprecation warnings. New docs and new code should use computation.

Conflict rule

Do not mix a non-auto computation value with enabled legacy flags.

# Raises ValueError.
pyg.walk(df, computation="browser", kernel_computation=True)

Use one expression of intent instead:

pyg.walk(df, computation="browser")

Static HTML limitations

Static HTML exports cannot call back into a live Python kernel or cloud service. The following are rejected:

pyg.to_html(df, computation="kernel")
pyg.to_html(df, computation="cloud")
pyg.to_html(df, kernel_computation=True)
pyg.to_html(df, cloud_computation=True)
pyg.to_html(df, use_kernel_calc=True)

Use browser computation for static exports:

html = pyg.to_html(df, spec_path="./gw_config.json", computation="browser")

Use Notebook APIs, Streamlit, Gradio, or a webserver session when you need live kernel or cloud computation.

Adapter notes

Adapter	Computation notes
`pyg.walk`, `pyg.render`, `pyg.table`	Accept `computation`; route to notebook adapters in Jupyter and webserver adapters elsewhere.
`pyg.Walker`	Stores computation mode for reuse. Static export methods still reject live kernel/cloud modes.
`pygwalker.api.anywidget.walk`	Defaults to kernel behavior when automatic mode is used in notebook widget contexts.
`StreamlitRenderer`	Defaults to kernel behavior when automatic mode is used; accepts `computation="browser"`, `"kernel"`, or `"cloud"`.
`get_html_on_gradio`	Accepts `computation` and live backend modes.
`pyg.to_html`	Browser/static only.
Reflex adapter	Rejects kernel and cloud computation; use `computation="browser"`.