DataCLI

DataCLI generates JSONL chat datasets and agent traces from structured JSONL prompts. It includes a command-line interface, a reusable TypeScript SDK, and a core execution runtime.

Requirements

• Node.js 22.19.0 or newer
• An API key for generate, or provider credentials supported by Pi for trace

Install

npm install --global @teichai/datacli

Quick start

Create prompts.jsonl:

{"prompt":"Explain the CAP theorem in simple terms.","metadata":{"topic":"distributed-systems"}}
{"prompt":["Write a JavaScript function that reverses a string.","Now add input validation."],"metadata":{"topic":"javascript"}}

Set an OpenRouter API key and generate a dataset:

export OPENROUTER_API_KEY="your-key"
datacli generate \
  --model openai/gpt-4o-mini \
  --prompts prompts.jsonl \
  --out dataset.jsonl

Each input record produces one validated output record. See Getting started for the complete walkthrough.

Documentation

• Documentation home
• Getting started
• CLI reference
• Configuration files
• Prompts and output format
• SDK and custom pipelines
• Providers
• Environments and agent traces
• Architecture and execution model
• Operations and troubleshooting
• Development

Development

npm install
npm test

See Development for workspace structure, build commands, and testing guidance.

License

Apache License 2.0