Analytics workspace¶

The analytics workspace is a Python-backed computational environment for data analysis. Upload a CSV, connect a database, or point at a file in your workspace - then ask the agent to analyse it, generate charts, and summarise findings.

How it works¶

The analytics workspace runs Python in a sandboxed subprocess. The agent has access to a pre-installed data science stack:

• pandas and polars for tabular data
• matplotlib, seaborn, and plotly for charts
• scipy and statsmodels for statistical analysis
• scikit-learn for machine learning
• duckdb for SQL queries over files and dataframes

Generated charts are returned as images embedded in the conversation. Code and outputs are shown in a collapsible code block.

Routes¶

Opening the analytics workspace¶

Navigate to /analytics or click the Analytics card in the launcher.

You can also start an analytics session from chat with /analytics followed by a question.

Uploading data¶

Drag a file onto the analytics workspace or use the upload button. Supported formats:

• CSV and TSV
• Excel (.xlsx, .xls)
• JSON and JSONL
• Parquet
• SQLite database files

Uploaded files are available to the agent as named dataframes. The agent identifies the schema automatically and summarises it before you ask any questions.

Connecting a database¶

Connect to a live database for direct querying:

KEPRIX_ANALYTICS_DB_CONNECTORS=postgres,mysql,sqlite

In the UI: Analytics > Connect database, select the type, and enter credentials. The connection is scoped to the analytics workspace session.

Supported databases: PostgreSQL, MySQL, SQLite, DuckDB, BigQuery, Snowflake.

Example queries¶

Just ask in natural language:

Show me the top 10 customers by revenue last quarter

Plot monthly sales as a bar chart, overlaid with a 3-month rolling average

Run a linear regression of temperature on ice cream sales

What percentage of orders have a discount applied?

The agent writes and runs the Python code, shows you the code and output, and provides a plain-language interpretation.

Code transparency¶

Every chart and analysis result shows the code used to produce it. Click Show code on any result to see the exact Python that ran. Copy, edit, and re-run it from the UI.

Sandboxing¶

Analytics code runs inside a restricted subprocess:

• File system access limited to the workspace data directory.
• Network access disabled (no outbound requests from analysis code).
• Memory and CPU limits enforced.
• Timeout: KEPRIX_ANALYTICS_TIMEOUT seconds (default 60).

Configuration¶

KEPRIX_ANALYTICS_ENABLED=true
KEPRIX_ANALYTICS_TIMEOUT=60          # seconds per code execution
KEPRIX_ANALYTICS_MAX_ROWS=1000000    # rows per uploaded file
KEPRIX_ANALYTICS_MAX_FILE_MB=100     # upload size limit

Saving and exporting results¶

• Save chart: click the download icon on any chart to save as PNG or SVG.
• Export notebook: export the full session as a Jupyter notebook (.ipynb) with all code cells and outputs.
• Export report: generate a Markdown or PDF report summarising the analysis.

API¶

POST /api/analytics/sessions                  # create a new analytics session
POST /api/analytics/sessions/{id}/upload      # upload a data file
POST /api/analytics/sessions/{id}/query       # run a natural-language query
GET  /api/analytics/sessions/{id}/results     # list results
GET  /api/analytics/sessions/{id}/export      # export as notebook

Related¶

• Deep research
• Self-coding agent
• Evals
• Playbooks
• RAG pipelines