A written word does not become what it names.

A name on a list.

Writing: My name is on the participant list. Does that mean this written entry is me?

After “My name”, the question asks whether “this written entry is me”. A name can be learned and recognized as referring to you. That step proposes treating the written name as the person it names. Even when read as yours, the name remains writing.

One of six prepared examples. Open the examples without a model request.

Your text and model connection

API requests use tokens from beforeword’s budget, so the website provides 8 analyses per browser without an account or personal key. The allowance does not renew daily; each separately submitted portion of an article uses one analysis when completed.

You can paste a long text in full and choose a portion to send. Its boundaries are visible; notes on that portion do not cover the rest of the article. Completed notes stay available while the panel is open and the source is unchanged. Returning to a completed portion makes no new model request.

The extension lets you choose a provider and model with an API key: OpenAI, Claude, Gemini, Grok, OpenRouter, DeepSeek, Mistral, Groq or a compatible API. The recipient is shown in the panel. The beforeword symbol beside a selection only opens the text in the panel. In the extension, “Examine writing” or Ctrl/Cmd+Enter in the text field sends the passage. On the website, text is sent through the analysis button. Selecting text or opening a prepared example makes no model request.

Before sending to your own model, the panel shows the selected account’s required data-processing mode. Text is not sent until the user confirms these checks; the extension does not independently verify account settings. OpenRouter requests require upstream routes classified as denying data collection and supporting ZDR; account logging and sharing remain separate. Gemini requires active billing; Mistral requires a reviewed GA model with improvement-data use disabled. Direct DeepSeek API remains listed with an explanation of its unavailable mode; a DeepSeek model can be chosen separately through an eligible OpenRouter route.

An analysis shows a proposed reading: how words are linked and what they are taken to describe. Such a link can be added while the letters stay the same. The explanation remains writing; it does not establish how you read the passage.

The dictionary is available separately: choosing a word loads its entry from Wiktionary. This is not AI generation and needs no account or API key. Loading an entry requires an internet connection.

Data and permissions

A dictionary request sends the chosen word or entered lookup form to Wiktionary (Wikimedia). That request does not send your whole source passage. Entries and your path stay in the open panel’s memory.

Introductory analysis sends the selected passage, or the chosen portion of pasted text, to beforeword and OpenAI. A personal connection sends it from the extension to the chosen provider; OpenRouter forwards it to the model supplier. Personal API keys are not sent to beforeword.

The extension requests access to chosen sites to capture text and show local highlights. It does not read other apps, passwords or chat cookies. The source and analysis stay in interface memory; the application server does not retain or log them. Hosting and provider processing follow their own policies.

Language, connection settings and a random installation identifier are stored for operation and the introductory allowance. A personal key stays in temporary session storage or extension memory. Forget key removes it; resetting the panel does not reset the allowance.

Extension data and connection requirements · Support: mail@beforeword.xyz