VVDex
Open the app
Menu

Quickstart · Web app

From first visit to a signed result

Everything here happens at vvdexops.com/app. No install, no credit card.

  1. Open the app

    No sign-in is needed: your first visit opens a private workspace kept in this browser. You can delete it any time under Settings.

  2. Connect a model

    Open Models, pick a provider and paste your key, or give an OpenAI-compatible address, or describe an HTTP app. Then run the check: it makes a live call and names anything wrong, such as a refused key, an unknown model or a reply in an unexpected shape.

  3. Compare answers in the playground

    Choose up to eight models, write one prompt, and read the answers side by side with latency, tokens and cost. A cost shows as unknown when the model’s price is not known; you can add prices to a saved model. Save the run and share it by link if you like.

  4. Run a Forge exam

    Open Forge exams, pick an exam and read what it asks and the checks it passed before it was offered. Choose models and a number of attempts; each attempt is a full working session in a fresh, locked sandbox. Progress shows live while the run works.

  5. Read the result

    Each model gets a pass rate with its 95% interval, errors by kind (kept out of the score), cost and latency. Open any case to see every model’s answer and the reason for each grade. Compare two runs to see regressions and improvements.

  6. Keep the receipt

    Download the run’s receipt, share the run by link, or publish a completed Forge exam run so anyone can see it. Anyone can check the receipt at /check-receipt.

Test on your own data

Open Datasets, create one and upload a CSV or JSONL file. The input column is the prompt (a string, or a list of messages in JSONL); expected is the answer to grade against; any other column is kept as metadata. Then start a run from Evaluations and choose graders: exact match, format, similarity, model-judged or safety.

A model-judged score is marked advisory until you have labelled enough answers by hand to measure how often the judge agrees with you.

Keys for code and CI

Under Settings → API keys, create a key for the API, SDK and CLI, GitHub Action or MCP. Keys can only be created from a signed-in browser, and each is shown once. Settings also holds your usage against the plan, the audit log and Delete account.