Data Q&A

Ask questions about local data files — row counts, totals, top values, summaries, and more. Query a single file, an entire folder, or only specific formats.

Natural language

arka ask data users.csv "how many rows?"
arka ask data ./data/ "what is total revenue?"
data_ask reports/ "summarize all files"
data_ask folder ./exports --question "how many records total?"
arka analyze csv files in data/ "top 10 products"
arka query_data sales.json "what is total revenue by category?"

CLI commands

Command	Description
`data_ask users.csv "how many rows?"`	Ask a question about a file
`data_ask ./data/ --format csv "average salary?"`	Query only CSV files in a folder
`data_ask ./reports/ --formats csv,json "list all columns"`	Filter by multiple formats
`data_ask folder ./exports --question "how many records total?"`	Explicit folder mode
`ask_data sales.json "total by category"`	Alias for `data_ask`
`query_data report.tsv`	Summarize when no question is given
`analyze_data data.jsonl "top 5 products"`	Alias for `data_ask`

View CSV as a colored table

For display (not Q&A), use view_data — inspired by tools like csvlook / pretty printers:

view_data papers.csv
view_csv papers.csv --plain
show csv papers.csv as json
cat data.tsv | view_data --formats csv,yaml

Colored columns print in the terminal; optional --formats writes exports under ~/arka-generated/ (override with DATA_OUTPUT_DIR). Analytic questions still route to data_ask.

Supported formats

csv, json, jsonl, tsv, yaml, yml, xml, parquet, xlsx, xls, sql Optional dependencies for richer parsing:

Format	Package
Parquet	`pyarrow` or `pandas`
Excel (xlsx)	`openpyxl` or `pandas`
YAML	`pyyaml`

Without pandas, tabular files still load via the Python standard library (csv, json).

Folders and format filters

When the path is a directory, Arka discovers data files by extension (non-recursive, top level only). Use --format or --formats to limit which types are loaded. Large folders are capped for context size:

Up to 5 files per query
Up to 500 rows total across loaded files
Per-file samples and column statistics when files are larger

How it works

Detects file format(s) from extension(s), optionally filtered by --format
Loads rows from one or more files (sample + stats when truncated)
Sends structured context to the LLM with your question
Prints the answer in the terminal

This skill is for querying existing files. To create fake sample datasets, use Generate data.

Examples

data_ask ~/Downloads/orders.csv "what are the top 3 products by quantity?"
data_ask ./data/ --format csv "average salary?"
query_data metrics.jsonl "average latency by region"
analyze_data inventory.parquet
data_ask reports/ "summarize all files"

Routing & CLI

Agent & memory

Developer tools

Models & inference

Media & documents

Integrations

Finance & data

Workflows & productivity

More

Natural language

CLI commands

View CSV as a colored table

Supported formats

Folders and format filters

How it works

Examples

​Natural language

​CLI commands

​View CSV as a colored table

​Supported formats

​Folders and format filters

​How it works

​Examples

Natural language

CLI commands

View CSV as a colored table

Supported formats

Folders and format filters

How it works

Examples