fabricatio-plot

Data visualization, synthetic data generation, and dataframe manipulation for the Fabricatio LLM agent framework.

Installation

pip install fabricatio[plot]
# or
uv pip install fabricatio[plot]

For optional file-format support:

pip install fabricatio[plot,excel]    # Excel I/O
pip install fabricatio[plot,parquet]  # Parquet I/O

Install with all Fabricatio components:

pip install fabricatio[full]

Overview

fabricatio-plot equips LLM agents with tools for matplotlib-based charting, programmatic synthetic data generation, dataframe CRUD operations, and file I/O. Toolboxes expose fine-grained functions agents can call; capabilities wrap them into higher-level LLM-driven workflows.

Key Components

Toolboxes

Low-level tools grouped by domain, exposed to agents via fabricatio-tool's ToolBox.

PlottingToolBox — matplotlib chart construction

DataSynToolbox — synthetic column generation

DataCrudToolbox — dataframe create/read/update/delete

DataIoToolBox — file-based I/O

Capabilities

Higher-level abstractions that orchestrate toolboxes with LLM reasoning.

• Plot — aggregates PlottingToolBox, DataCrudToolbox, and DataIoToolBox. The plot(requirement, data, output_spec) method accepts a natural-language requirement and delegates to its toolboxes to produce charts.
• SynthesizeData — uses LLM to generate structured data from a natural-language description. generate_header() produces column names; generate_csv_data() returns a DataFrame; synthesize_data() handles large datasets by batching parallel generation.

Actions

Concrete fabricatio-core Action implementations ready to wire into task pipelines.

• MakeCharts — takes a plot requirement (or the task's assembled prompt) and produces a chart, optionally saving to a specified path.
• MakeSynthesizedData — synthesizes a DataFrame from the task prompt and stores it under the key synthesized_data.
• SaveDataCSV — saves a DataFrame to a path (given explicitly or determined by the LLM from the task context).

Usage

Toolboxes are stateless and can be called directly:

from fabricatio_plot.toolboxes.plot import create_figure, plot_line, set_labels, save_plot

fig = create_figure(figsize=(10, 4))
ax = fig.subplots()
plot_line(ax, x=[1, 2, 3], y=[4, 5, 6], label="Series A")
set_labels(ax, title="Sample Chart", xlabel="X", ylabel="Y")
save_plot(fig, "output.png")

Capabilities drive LLM-orchestrated workflows:

from fabricatio_plot.capabilities.plot import Plot

handler = Plot()
result = await handler.plot("Create a bar chart of monthly sales from the CSV at data/sales.csv")

Actions plug into Fabricatio task graphs:

from fabricatio_plot.actions.plot import MakeCharts

action = MakeCharts(plot_requirement="Scatter plot of x vs y", chart_save_path="scatter.png")
await action.execute(task)

Dependencies

• fabricatio-core — core interfaces, configuration, and task model
• fabricatio-tool — ToolBox and Handle abstractions
• matplotlib — chart rendering
• numpy — array operations
• pandas — DataFrame manipulation
• Optional: openpyxl (Excel I/O), pyarrow (Parquet I/O)

Configuration

All options below are read through the fabricatio configuration chain (see the Configuration Guide). Set them under the [ext.plot] table in fabricatio.toml, equivalently under [tool.fabricatio.ext.plot] in pyproject.toml, or via FABRICATIO_EXT__PLOT__<FIELD_UPPER> environment variables.

[ext.plot]

Access at runtime: from fabricatio_plot.config import plot_config.

License

MIT — see LICENSE