SchemaForge
source: Structured Data with LLMs Done Right · by: Jesus Villota
LLM Provider
OpenAI (GPT)
GPT-4o, GPT-4o-mini
GroqCloud
GPT-OSS 120B, Kimi K2, Llama 4 Scout
Anthropic (Claude)
Claude 4.5 & 4.6 (Opus, Sonnet, Haiku)
Google (Gemini)
Gemini 3.1 Pro, Gemini 2.5, 2.0 Flash
Mistral
Mistral Large, Small
DeepSeek
DeepSeek Chat, Reasoner
Meta (Llama)
Llama 4 Maverick, Llama 3.3
xAI (Grok)
Grok 4.1 Fast, Grok 4 Fast
Model
Loaded example: firm-level shock parser from Villota (2024) — edit freely or to start from scratch.
Schema name
Free-form text or constrained to specific values (enum). Great for classifications and open-ended text.
Floating-point decimal numbers (e.g., 3.14, 0.5). Use for measurements, percentages, and continuous values.
Whole numbers only (e.g., 42, -5, 0). Use for counts, indices, and discrete quantities.
True or false only. Use for yes/no decisions, presence/absence, or binary classifications.
Ordered list of items (strings, numbers, objects, etc.). Use for multiple related values like tags or nested records.
Nested structure with its own fields/properties. Use for complex data with multiple related attributes.
🧱 Your schema is empty

Pick a type above to add your first field. You can combine string, number, integer, boolean, array, and object to build any schema you need.

System prompt
User prompt template
Generation parameters
0
Live Preview

Add fields in Step 2 and the generated code will appear here in real time.