Frequently asked questions
Short, honest answers to the questions people actually ask about StrictJSON. If yours is not here, the tool itself is the fastest way to find out — describe what you need and generate.
Is my API key stored?
No. StrictJSON works without any key for the free tier. If you add your own OpenRouter key (BYOK — bring your own key), it is stored only in your browser's localStorage, on your own machine. It is sent to OpenRouter — and nowhere else — only when you click Generate. On the server side, no visitor key is ever stored.
Which models generate the schemas?
Four models, chained server-side with automatic verification: Gemini gemini-3.8-flash first, then Groq openai/gpt-oss-120b, then OpenRouter nvidia/nemotron-3-ultra:free, then google/gemma-3-4b-it:free. Each step checks the previous one's output; if a step fails, the chain falls through to the next model. The chain exists so that one model's bad day does not become your bad output: a schema is only returned once it has passed the server-side quality check, which also repairs common mistakes automatically.
What is the daily limit?
The free tier allows 10 generations per day, per IP address. If you hit the limit, you can keep going with your own OpenRouter key — paste it in the tool, and it is used only from your browser, as described above.
Are the examples tested?
Yes — for real, by execution, not by reading. Every library entry is run through its Pydantic model, its Zod schema, and its JSON Schema validator, and the results are checked. An example only ships when all three pass. See the library for the twelve tested entries.
Can I use the output in production?
Yes, with one condition: always validate server-side. A schema constrains the shape of the model's output — it cannot stop the model from inventing a plausible-looking value inside that shape. Treat generated output like any untrusted input: run it through Zod, Pydantic, or a JSON Schema validator in your own code before it touches your database. Read Why Strict JSON? for the full reasoning.
What if validation fails?
First, check your data: most failures are a value that does not match the pattern, a missing required field, or an extra key when the schema is strict. Read the validator's error message — it names the exact field and the exact rule that failed, which is usually enough to see whether the data or the schema is at fault. If the data looks right, regenerate — the same description can produce a slightly different schema, and a second generation often fixes it. If the problem persists, tighten the description: name the exact fields, the exact formats, and what must never be invented. A vague description produces a vague schema, and a vague schema is what validation exists to catch.
Do I need an account?
No. There is no sign-up, no login, and no tracking account. Open the tool, describe what you need, generate, and copy. The free daily limit is counted per IP address, not per user.