kedro-mermaid lets you focus diagrams on the slices of the pipeline that matter. All filters stack, so you can combine datasets, nodes, namespaces, and tags to tighten the view.

Target a Specific Pipeline

Pass the pipeline name with --pipeline (defaults to __default__):

kedro mermaid generate --pipeline=data_science

If the name is invalid the command reports the list of registered pipelines so you can pick the right one.

Trim by Datasets

Use --from-inputs and --to-outputs to cut the diagram around datasets:

kedro mermaid generate --from-inputs=raw_customers,raw_products --to-outputs=model_predictions

• --from-inputs keeps only paths reachable from the listed datasets.
• --to-outputs keeps only paths that lead to the listed outputs.

Trim by Nodes

When you need to focus on specific Kedro nodes, you have three options:

kedro mermaid generate \
  --from-nodes=data_preprocessing.clean_customers \
  --to-nodes=model_training.train_model \
  --nodes=model_training.evaluate_model

• --from-nodes starts traversal from selected nodes.
• --to-nodes stops traversal when a listed node is reached.
• --nodes only includes the nodes you list, collapsing edges through any missing nodes.

Filter by Tags or Namespaces

Kedro metadata can scope the diagram too:

kedro mermaid generate --tags=model,reporting
kedro mermaid generate --namespaces=data_science.pipelines

• --tags keeps nodes that carry the specified Kedro tags.
• --namespaces keeps nodes inside selected namespaces. On Kedro 0.x projects the plugin transparently falls back to the legacy node_namespace argument.

Filter by Regex Pattern

Regex parsing acts as another filter layer via --set-node-attr pattern=<regex>:

kedro mermaid generate \
  --set-node-attr pattern="(?P<category>\w+)__(?P<node>\w+)__?(?P<node>\w+)?"

When the pattern matches a node name:

• Named captures labelled node (or numbered groups) populate the node label. Multiple captures are joined with spaces and capitalised.
• A named capture labelled category groups nodes into a subgraph. Each category appears as a coloured box with autoregistered styling.

Nodes that do not match the pattern are hidden from the final diagram, but the graph traverser still uses them to maintain continuity between any nodes that stay visible. Skip the pattern or make it permissive if you want every node to remain.

Understand Simplification

After category and pattern parsing, the plugin simplifies the graph:

1. Nodes that fail filters (including regex patterns) are removed.
2. Direct paths are re-linked so you never see dangling intermediates when focusing on a subset.
3. Self-loops are dropped automatically.

A simple example: A -> B -> C becomes A -> C when only A and C match the filters.

You can inspect the simplified edges by checking the output markup or sending it to a file:

kedro mermaid generate > pipeline.mmd

From there, open the file in the Mermaid Live Editor or embed it in your docs pipeline.

Combine Filters

Filters combine, so you can stack criteria:

kedro mermaid generate \
  --pipeline=pipelines.data_science \
  --from-inputs=raw_transactions \
  --to-outputs=reporting.kpis \
  --tags=reporting

The command above follows the reporting path within the data_science pipeline, starting from the raw transactions dataset, ending at the KPI report, and keeping only reporting-tagged nodes.

When an overly strict combination yields an empty result, kedro-mermaid prints a simplified diagram with no edges. Loosen the filters until the desired section appears.