maestro-analyze 0.2.0

Analyze everything, for agents. Personal data analysis engine with DuckDB.

maestro-analyze

Analyze everything, for agents.

Import any dataset, profile it, query it with SQL, and generate 25+ chart types -- all from the command line. DuckDB backend. Zero config. Built for AI agents and humans who prefer terminals over dashboards.

---

Quickstart

pip install maestro-analyze

$ manalyze import sales.csv
  Imported: sales (12,847 rows, 9 columns)

$ manalyze profile sales
  sales -- 12,847 rows x 9 columns
  Revenue is right-skewed (median $420, mean $1,230)
  3 outliers detected in profit_margin (> 3 sigma)
  Missing: 2.1% in region column

$ manalyze chart sales --type bar --x region --y revenue
  Chart saved: charts/sales.html

$ manalyze query "SELECT region, SUM(revenue) FROM sales GROUP BY region ORDER BY 2 DESC"

---

Features

• DuckDB backend -- import CSV, Excel, JSON, Parquet into a local analytical database. SQL everything.
• Auto-profiling -- descriptive stats, distribution analysis, anomaly detection, missing data audit.
• 25 chart types -- interactive Plotly charts and pure SVG charts. Auto-selects type when you omit --type.
• Trend analysis -- time series aggregation with configurable frequency (daily, weekly, monthly, quarterly, yearly).
• Comparison analysis -- group-by metrics with statistical insights.
• Plugin architecture -- drop a .py file in ~/.maestro/analyst/plugins/charts/ and it auto-registers.
• Agent-native -- every command returns structured output. Designed as a tool for AI coding agents.

---

Chart Types

Plotly (interactive HTML/PNG/SVG)

Type	Description
bar	Bar chart
line	Line chart
scatter	Scatter plot
histogram	Histogram
pie	Pie chart
box	Box plot
bubble	Bubble chart (size + color dimensions)
sankey	Sankey flow diagram
treemap	Hierarchical treemap
wordcloud	Word cloud
radar	Radar / spider chart
heatmap	Correlation heatmap
distribution	Distribution histogram with KDE
funnel	Funnel chart
table	Formatted data table
event_band	Line chart with event period bands
stacked_area	Stacked area composition
bland_altman	Agreement / validation plot
coverage_matrix	Data completeness heatmap
trust_radar	Multi-axis scoring radar

SVG (pure, no JavaScript dependency)

Type	Description
bump	Ranking trajectories over time
heatmap_grid	2D categorical grid heatmap
lollipop	Horizontal lollipop chart
event_timeline	Vertical event timeline
slope	Period comparison slope chart
sparkline	Tiny inline chart

---

CLI Reference

Import data

manalyze import data.csv                          # auto-detect table name from filename
manalyze import data.xlsx --table quarterly        # custom table name
manalyze import data.parquet --db project.duckdb   # custom database file

List tables

manalyze tables

Profile a table

manalyze profile sales                            # auto-insights + descriptive stats

SQL queries

manalyze query "SELECT * FROM sales WHERE revenue > 1000" --limit 100

Generate charts

manalyze chart sales --type bar --x region --y revenue
manalyze chart sales --type scatter --x cost --y revenue --color region
manalyze chart sales --type bubble --x cost --y revenue --size volume --color region
manalyze chart sales --type sankey --source origin --target destination --values volume
manalyze chart sales --type treemap --names category --parents parent --values revenue
manalyze chart sales --type bump --x year --y rank --color entity
manalyze chart sales --sql "SELECT region, SUM(revenue) r FROM sales GROUP BY 1" --type bar --x region --y r
manalyze chart sales --type line --format png      # export as PNG instead of HTML
manalyze chart sales --type bar --fragment          # output <div> fragment for embedding

Trend analysis

manalyze trend sales --date order_date --metric revenue              # monthly by default
manalyze trend sales --date order_date --metric revenue --freq W     # weekly

Comparison analysis

manalyze compare sales --group region --metrics "revenue,profit,volume"

Export

manalyze export sales                              # CSV by default
manalyze export sales --format parquet             # Parquet
manalyze export sales --format json -o out.json    # JSON to specific path

Drop a table

manalyze drop old_data

---

Python SDK

from maestro_analyze.core.store import Store
from maestro_analyze.engine.analyzer import Analyzer
from maestro_analyze.engine.charts import make_chart, save_chart

# Import and query
with Store("workspace.duckdb") as store:
    store.import_file("sales.csv")
    df = store.query("SELECT region, SUM(revenue) as rev FROM sales GROUP BY 1")

# Auto-profile
with Store("workspace.duckdb") as store:
    analyzer = Analyzer(store)
    result = analyzer.profile("sales")
    for insight in result.insights:
        print(f"{insight.title}: {insight.detail}")

# Generate charts programmatically
fig = make_chart(df, "bar", x="region", y="rev", title="Revenue by Region")
path = save_chart(fig, "revenue_by_region", "html")

---

Architecture

CLI (typer)  -->  Store (DuckDB)  -->  Analyzer (profiling, trend, compare)
                                  -->  Chart Engine (auto-select or manual)
                                         |
                                  chart_builders/
                                    __init__.py        # auto-discovers via pkgutil
                                    _base.py           # BaseChartBuilder ABC
                                    _svg_base.py       # SVG helpers
                                    bar.py             # one file per chart type
                                    ...                # 25 built-in types
                                  ~/.maestro/analyst/plugins/charts/
                                    my_chart.py        # user plugins (auto-registered)

Adding a custom chart

Drop a .py file in ~/.maestro/analyst/plugins/charts/:

from maestro_analyze.engine.chart_builders._base import BaseChartBuilder

class WaterfallBuilder(BaseChartBuilder):
    name = "waterfall"
    description = "Waterfall chart for incremental changes"

    def build(self, *, df, **kw):
        import plotly.graph_objects as go
        fig = go.Figure(go.Waterfall(x=df[kw["x"]], y=df[kw["y"]]))
        return fig

Available on next run -- no registration code needed.

---

How it compares

	maestro-analyze	Metabase	Jupyter	Apache Superset
Setup	pip install	Docker + database	pip install + kernel	Docker + database + Redis
Interface	CLI + Python SDK	Web UI	Notebook	Web UI
Agent-friendly	Yes (structured CLI output)	No (browser-only)	Partial (kernel protocol)	No (browser-only)
Chart types	25 built-in + plugins	~15	Unlimited (matplotlib)	~30
Data backend	DuckDB (embedded)	PostgreSQL / MySQL	pandas (in-memory)	SQLAlchemy
Auto-profiling	Built-in	No	Manual	No
Config required	None	Database, SMTP, secrets	None	Database, cache, auth
Target user	Agents + CLI developers	Business analysts	Data scientists	BI teams

maestro-analyze is not a replacement for full BI platforms. It is a zero-config analysis tool designed to be called by AI agents or used directly from the terminal.

---

Optional extras

pip install maestro-analyze[ml]       # scikit-learn, scipy (clustering, statistical tests)
pip install maestro-analyze[polars]   # Polars DataFrame support
pip install maestro-analyze[all]      # everything

Development setup

git clone https://github.com/maestro-ai-stack/maestro-analyze.git
cd maestro-analyze
python3.11 -m venv .venv && source .venv/bin/activate
pip install -e ".[dev]"
pytest tests/ -v

---

Contributing

Core improvements -- open issues and PRs on maestro-ai-stack/maestro-analyze.

Custom chart types -- contribute built-in builders via PR, or distribute your own as plugin files that users drop into ~/.maestro/analyst/plugins/charts/.

---

License

MIT

---

Built by Maestro -- Singapore AI product studio.