cutl - Your friendly jq-based JSONL Dataset Editor

cutl is a fast, interactive TUI tool to view, filter, and edit JSONL (JSON Lines) files – ideal for NLP and machine learning datasets, or any structured line-oriented data. It's goal is to be pleasant to use. The primary intended use-case is to work in NLP dataset / corpus files that need to be refined, filtered or edited. (e.g. the ones created for https://spacy.io)

Most cutl operations—including filtering, column selection, and row manipulation—are powered by jq selectors and syntax. Users familiar with jq will find expressive power for querying and editing structured data, and all filtering follows jq-compatible rules. For a primer, see the jq manual.

Features

• Interactive table view for large JSONL files
• Live filtering and JQ-style queries
• Optional AI-assisted filter prompts (requires OPENAI_API_KEY or ANTHROPIC_API_KEY)
• Easy field/row editing, supports multi-line edit
• Keyboard-friendly navigation (vim- and arrow keys)
• Batch delete, mark/clear, save back to file
• Detail and column configuration views
• Works anywhere Go runs (no runtime dependencies)

Quick Start

git clone <repository-url>
cd cutl
go mod tidy
go build -o cutl
./cutl --input path/to/data.jsonl

Usage Example

./cutl --input data.jsonl         # Open and edit

For keyboard shortcuts, see in-app help.

AI-assisted filtering

If you export OPENAI_API_KEY or ANTHROPIC_API_KEY, cutl unlocks an “AI Filter” prompt that can turn natural language instructions into jq filters.

OpenAI:

export OPENAI_API_KEY=sk-your-key
# Optional overrides
export CUTL_OPENAI_MODEL=gpt-4.1-mini
export OPENAI_BASE_URL=https://api.openai.com/v1

Claude (Anthropic):

export ANTHROPIC_API_KEY=sk-ant-your-key
# Optional override (defaults to claude-opus-4-8)
export CUTL_ANTHROPIC_MODEL=claude-opus-4-8

If both OPENAI_API_KEY and ANTHROPIC_API_KEY are set, cutl defaults to OpenAI; set CUTL_AI_PROVIDER=anthropic (or openai) to choose explicitly.

While viewing the table, press P to open the AI prompt. Describe what you want to filter (e.g. “rows where language is German and score > 0.8”) and press Enter. The currently selected row is sent as context so the model understands your data structure. The assistant responds with a jq expression that is immediately applied as the active filter. If no API key is available, the shortcut is hidden and the regular filtering workflow stays unchanged.

Contributing

Pull requests are welcome. The tool is heavily vibe-coded up to this point, so don't expect the finest of code.

License

MIT