JSON Schema Generator for LLM Structured Outputs

A JSON schema describes the exact shape of the data you want a language model to return: field names, types, which fields are required, and what each field means. OpenAI, Claude, and Gemini all accept a schema alongside your prompt and constrain the model's response to match it, which turns free-form text generation into reliable, parseable data extraction.

OpenAI's structured outputs strict mode requires additionalProperties: false on every object and every property listed in the required array. Because of that, a field can't simply be left out of required to make it optional. Instead its type becomes a union with null, like ["string", "null"], and the model returns null when the field doesn't apply. This tool applies those transforms automatically on the OpenAI tab.

OpenAI takes a response_format block with a strict JSON schema and the transforms strict mode demands. Claude takes the schema as a tool definition's input_schema and treats optional fields normally: they're omitted from required. Gemini accepts standard JSON Schema through the responseJsonSchema field of generationConfig. The schema you build here is converted to each provider's dialect automatically.

Toggle the field to optional in the builder. On the OpenAI tab the field stays in the required array (strict mode demands it) but its type becomes a null union, so the model returns null instead of omitting the key. On the Claude and Gemini tabs the same field is simply left out of required.

Yes. Paste a sample of the JSON you want on the builder's empty state and the tool infers field names, types, optionality, and nesting, including merging the elements of arrays. Then switch to the Zod tab for z.object() code with .describe() carrying your field descriptions, ready for the Vercel AI SDK's generateObject or streamObject.

No. Everything runs in your browser. The share link encodes the schema in the URL fragment (the part after #), which browsers never send to the server.