THE DORK MANUAL

Providers and models

GPT-6.1 Sol is Dork’s planned launch default. Current Labs are restricted to Luna, so Sol must not be treated as available until your release and Lab enable it. Check the model offered by your connected Lab before starting work. A model mentioned on this website is not evidence that your account can run it.

ChatGPT subscription

The hosted subscription path uses your own ChatGPT sign-in through Codex. You authorize it on the provider’s sign-in page; you do not paste a ChatGPT password into a research prompt. Your plan’s model access and usage limits still apply.

An OpenAI API key is a separate connection with separate billing and data terms. Connecting ChatGPT does not fund API calls, and adding an API key does not prove a ChatGPT plan is connected.

Claude subscription

Claude subscriptions and Anthropic API access are also separate. Claude subscription access is currently limited to development setups. The hosted configuration disables that access; a Claude data-use disclosure in the app does not establish support. Do not assume a Claude subscription can run hosted Dork research. Use only connections your release and Lab explicitly support.

API connections

Where supported, Provider connections lists OpenAI, Anthropic, and Google Gemini. Only the Lab owner can change these connections, and protected Lab storage is required. The desktop action asks for a private text file containing the provider key. Keep that file out of shared folders, research sources, and feedback attachments.

Connected describes a saved connection. It does not guarantee that every model is enabled, funded, or permitted for research. Refresh the model list after changing a key. If the list cannot be read, resolve that state before relying on it.

Data and review

Models receive the prompts and permitted context sent to them. ChatGPT, Claude, API, and custom-provider data controls differ. An OpenAI-compatible endpoint does not inherit OpenAI’s policies. Read the account-specific terms in Data use and retention.

Two models from the same provider family do not establish independent cross-family review. Judge the review disclosure and the evidence scope, rather than the number of researcher names.