xeli

Excel for the terminal. An interactive TUI spreadsheet with natural language queries.

A Really Artificial project.

Open CSV, JSON, Parquet, and Excel files in a fast TUI spreadsheet. Filter, sort, pivot, and export visually. Ask questions in plain English and let xeli write the SQL.

Try it without installing

→ www.reallyartificial.org/work/xeli/playground/

DuckDB-WASM in your browser. Load a CSV and query it locally with SQL. Optional AI queries send your question and column schema to your chosen provider using your API key.

Install

# npm (one-off, no install)
npx @josharsh/xeli examples/employees.csv

# npm (global)
npm install -g @josharsh/xeli

# Homebrew
brew install josharsh/tap/xeli

The npm package and Homebrew tap retain their existing names for compatibility.

Or grab a prebuilt binary.

Supported on macOS (Apple Silicon, Intel) and Linux (arm64, x86_64). Windows isn't supported yet, see #help-wanted.

Quick start

xeli sales.csv             # open a CSV
xeli data.parquet          # or Parquet, JSON, JSONL, Excel
cat events.json | xeli     # pipe anything
psql -c "..." | xeli       # query results, browsable
xeli                       # no args opens a file picker for the current dir

Press Ctrl+K and ask something:

• "top 10 employees by salary"
• "average revenue by city for active customers"
• "signups per week last quarter"

xeli sends your schema plus the question to OpenAI or Anthropic, runs the returned SQL on DuckDB, and shows the rows.

Features

Multi-format	CSV, TSV, JSON, JSONL, Parquet, xlsx
AI queries	Ctrl+K, plain English to SQL via OpenAI or Anthropic
SQL mode	Ctrl+Q, direct DuckDB queries with history
Filter	f, visual builder with 12 operators (=, >, contains, regex, ...)
Sort	s, cycle current column ASC / DESC / off
Search	/, regex across all columns, n and N to walk matches
Group-by + pivot	g, multi-column grouping wizard with COUNT, SUM, AVG, MIN, MAX
Join	J, pick a second file, choose INNER / LEFT / RIGHT / FULL
Computed columns	c, derive new columns from expressions (price * qty)
Formula bar	=, one-shot expressions (SUM(amount), MEDIAN(salary))
Histograms	v, sparkline distribution for any numeric column
Column stats	Ctrl+I, min / max / mean / median / nulls / unique count
Export	e, write filtered result back out as CSV / JSON / Parquet
Themes	t, Dracula, Nord, Catppuccin, Tokyo Night, Solarized
Command palette	Ctrl+P, fuzzy-search every command
Undo	u, full view-state stack
File picker	run xeli with no args for a fuzzy list of supported files in cwd
Mouse	click, scroll, drag to resize columns

Full keybinding list lives in the in-app help (?) or in src/handlers/input.rs.

How is this different from VisiData / csvkit / xsv / sc-im?

They're all great. Here's how xeli finds its niche:

Tool	Best for	What xeli does differently
VisiData	Mature, scriptable Python TUI	DuckDB engine plus an AI natural-language query layer
csvkit	Pipeline transforms	Interactive browsing instead of one-shot scripts
xsv	Blazing fast CSV processing	Multi-format (Parquet, JSON, xlsx) plus TUI plus SQL
Miller	Streaming transforms across formats	Interactive table, group-by wizard, joins by point-and-click
sc-im	Classic spreadsheet UX	Less Excel-cell-editing, more "browse + query large data"
q	SQL over CSVs from the shell	Same idea, plus a TUI, plus NL queries, plus more file formats

If you live in a notebook for ad-hoc data exploration, xeli is the "I just want to look at this CSV without spinning up Python" tool.

AI setup

xeli works without AI. Every feature except Ctrl+K runs offline. To turn natural language queries on, set an API key:

# OpenAI
export OPENAI_API_KEY=sk-...
# or: xeli config set-key openai sk-...

# Anthropic (Claude)
export ANTHROPIC_API_KEY=sk-ant-...
# or: xeli config set-key anthropic sk-ant-...

If you press Ctrl+K with no key set, xeli walks you through it inline. Pick a provider, paste your key, hit Enter. No restart needed.

Examples

Sample datasets live in examples/:

xeli examples/employees.csv      # 30 rows x 7 columns
xeli examples/departments.csv    # join target
xeli examples/employees.json     # same data, JSON

Try this in the AI bar (Ctrl+K):

• top 3 highest-paid people in engineering
• average salary by department, ordered by amount
• who has been here the longest in each city?

Or join two files (J):

1. Open examples/employees.csv
2. Press J, enter examples/departments.csv
3. Pick INNER join, both on the department column

Building from source

cargo build --release
./target/release/xeli examples/employees.csv

Tests (release mode is required on macOS):

cargo test --release

Tech stack

• Rust. Single static binary, ~3k LOC, no runtime dependencies.
• Ratatui for the terminal UI.
• DuckDB as the embedded vectorized SQL engine.
• clap for CLI parsing.
• tokio as the async runtime for the one AI request.

Demo

A short, deterministic walk-through. Open a CSV, navigate, sort by salary, then run the group-by wizard for AVG(salary) per department.

Contributing

PRs welcome. See CONTRIBUTING.md for the build and test loop and the open help-wanted list (Windows, Ollama backend, streaming, themes).

License

MIT. See LICENSE.