Today we’re excited to announce the launch of prompt2json v0.1.0, a Unix-style command-line tool that transforms natural language prompts into schema-validated JSON output. Released on December 20, 2025, this initial release brings production-ready capabilities for integrating LLM-powered data extraction directly into shell scripts and automation pipelines.

What’s New

As the inaugural release, v0.1.0 establishes the core functionality that makes prompt2json a reliable tool for LLM interactions in automated workflows:

Schema-Validated JSON Output

The tool validates all LLM responses against user-provided JSON schemas before outputting results. The --pretty-print flag provides human-readable formatting, while validation failures trigger exit code 4 for easy detection in shell scripts.

Multi-Provider Support

Works seamlessly with both Google Cloud’s Vertex AI (Gemini models) and any OpenAI-compatible endpoint. This flexibility lets you use commercial APIs, self-hosted models via Ollama, or other compatible services from a single CLI interface.

Pipeline-Friendly Design

Built to follow Unix conventions—valid JSON goes to STDOUT, diagnostics go to STDERR, and meaningful exit codes enable reliable pipeline composition with tools like jq, grep, and awk.

Configurable Timeout Protection

The --timeout flag lets you set HTTP request timeouts (default: 60 seconds), preventing hung operations in automated workflows.

Enhanced Diagnostics

Verbose mode provides detailed insight into input sources, schema compilation status, token usage, and validation results—without contaminating machine-readable output.

Why It Matters

Integrating LLMs into shell scripts has historically been challenging. Raw LLM responses are often unreliable for downstream processing, requiring brittle regex parsing or complex post-processing logic. prompt2json solves this by:

  • Enforcing structure: JSON Schema validation ensures outputs match expected formats before they reach your scripts
  • Enabling automation: Clean STDOUT/STDERR separation means pipelines work predictably
  • Reducing friction: Single-command interface replaces custom Python or Node.js glue code
  • Supporting experimentation: Repeatable, inspectable prompt workflows from the command line

This release represents a significant step toward treating LLM calls as deterministic interfaces rather than interactive sessions—making them viable for CI/CD pipelines, data processing jobs, and repeatable automation tasks.

Getting Started

Installation

Install directly from source using Go:

go install github.com/UnitVectorY-Labs/prompt2json@latest

Or download pre-built binaries from the v0.1.0 release page. Assets are available for Windows, macOS (Intel and Apple Silicon), and Linux across multiple architectures.

Quick Example

Here’s a simple example extracting structured data from text:

echo "Review this product feedback: Great service but slow delivery" | \
  prompt2json \
  --system-instruction "Extract sentiment and key topics" \
  --schema '{"type":"object","properties":{"sentiment":{"type":"string"},"topics":{"type":"array","items":{"type":"string"}}}}' \
  --project your-gcp-project \
  --location us-central1 \
  --model gemini-2.5-flash \
  --pretty-print

Authentication

For Gemini (Vertex AI), authenticate using Google Application Default Credentials:

gcloud auth application-default login

Or set a service account key:

export GOOGLE_APPLICATION_CREDENTIALS=/path/to/key.json

For OpenAI-compatible providers, supply your API key via --api-key or the OPENAI_API_KEY environment variable.

A Note on AI Assistance

This release announcement was drafted using an AI model (unsloth/Qwen3.5-122B-A10B-GGUF:Q4_K_M) to help generate user-facing documentation. The code and features described here were developed through collaboration between human contributors and GitHub Copilot coding agents. For full transparency, this post was AI-generated on March 17, 2026, by release-storyteller.