This tool is a static HTML file that runs entirely in your browser. There is no Extrafemi Labs server, no database, and no logging of any kind. Your conversations, your provocation, your framework — none of it touches any infrastructure Extrafemi Labs controls. We cannot see it, store it, or share it because we never receive it.
When you send a message, your text is transmitted to whichever provider you selected — Anthropic or OpenRouter — to generate the sparring partner's response. This is unavoidable given how the tool works. Neither provider uses API-tier data to train models by default. This protection applies specifically to the API key approach used here — not to consumer products like Claude.ai or ChatGPT, which operate under different terms.
On by default. The moment a debate reaches the Literature Debate stage, the provocation and domain are sent to Semantic Scholar, PubMed, Europe PMC, and DataCite — four free, public research databases — to check for existing work before the sparring partner challenges the framework's novelty. This is disclosed in the chat itself when it happens, not only here, and can be turned off from Data & Privacy inside the tool.
Every "Confident"-tier citation the sparring partner produces is automatically cross-referenced against CrossRef and OpenAlex as soon as it appears, with the result shown directly under that message. This always runs — it isn't optional, because a fabrication check that only happens when someone remembers to click a button can miss the exact thing it exists to catch. A "found" result confirms a real paper by that author in that year exists, not that the specific claim attached to it is accurate. Always verify independently before academic use.
Your API key is held only in your browser's memory for the current session. It is passed directly to the provider in each request and is never sent to any Extrafemi Labs server. It is not written to localStorage, cookies, or any persistent storage. When you close the tab, it is gone.
If you use the session restore feature, your debate transcript and framework notes are saved to your browser's localStorage — a storage mechanism that exists only on your device. This data never leaves your browser. Extrafemi Labs cannot access it. Clearing your browser data will delete it.
If you choose to enter your ORCID iD, your profile's public name and affiliation are looked up automatically from ORCID's public API and, together with your iD (and/or a ROR institution you search for separately), stored only in this browser's localStorage and used only to add one attribution line to records you export. None of it is transmitted anywhere else, and providing any of it is entirely voluntary — leaving these fields blank changes nothing else about the tool.
This site is hosted on Netlify, which may run its own built-in Real User Monitoring (Core Web Vitals — page-load performance only) at the infrastructure level. This tool sets no cookies of its own and runs no third-party analytics or advertising scripts. See Netlify's Privacy Policy ↗ for what that monitoring covers.
Because everything above is stored only on your own device and nothing is stored on any server we operate, most data-protection rights are already fully in your hands without needing to ask us:
Access & portability — open your browser's developer tools (localStorage) at any time, or use Export to get your full record in Markdown, Text, Google Docs, LaTeX, BibTeX, or RIS.
Rectification — edit or overwrite any field (domain, ORCID iD, institution, saved project name) directly in the tool at any time.
Erasure — clear this browser's site data, or use the "Clear all my data" button in the tool itself to wipe everything it has stored on this device in one step.
Objection / restriction — simply don't enter data you don't want processed (e.g. leave the ORCID/institution fields blank); nothing is collected passively.
Extrafemi Labs publishes this tool but, as described above, never receives or processes the content you generate with it — that goes directly from your browser to your chosen provider (Anthropic or OpenRouter) for every reply, automatically to CrossRef and OpenAlex for every confident-tier citation check, automatically to Semantic Scholar, PubMed, Europe PMC, and DataCite once per session at the Literature Debate stage (unless turned off in the tool), and to ORCID/ROR/Zenodo only for the specific optional features that call them. Those providers are based outside the EU/EEA in some cases; consult their own privacy policies (linked throughout this page) for their specific transfer safeguards. For questions about this policy, contact [email protected].