Disclaimer

What this site is and is not

worldthought.com is a reader’s aid built around public-domain primary-source philosophical texts. The on-page chatbots are AI systems that retrieve passages from the corpus and produce scholarly summaries with citations. They are not a substitute for reading the philosophers themselves, and they will occasionally make mistakes: misattribute a passage, conflate a position, or paraphrase in a way that loses a distinction the original text preserves. Verify anything important against the cited passage.

The chat reads from the texts listed in each room’s provenance README and does not have access to scholarship published after the language models’ training cutoffs except where the corpus has been extended with additional public-domain material.

Roadmap and corpus constraints

The Philosophers panel in the page header shows every study room currently on the site. New rooms open when a viable public-domain or freely-shared corpus is identified, ingested into Pinecone, and given a per-room prompt module. Several rooms are live with a constrained corpus because the standard twentieth-century English translations remain in copyright (Heidegger, Wittgenstein, Kierkegaard's Hong-Hong editions, and most of the late-twentieth-century continental and Anglo-American rooms): the chat in each of these rooms discloses the corpus thinness and reasons from training-data knowledge for the in-copyright material. The corpus widens for these rooms as older translations enter the public domain or out-of-copyright alternatives are identified.

Data and privacy

worldthought.com does not set cookies, does not run any third-party analytics, and does not embed any cross-site tracking. The site is a static React build served by Netlify plus one Netlify Function that streams the chat completions.

Two pieces of data are stored on Netlify Blobs in the course of running the chat:

• Rate limit state. One blob per IP address with the current minute’s request count. Reset by the next minute window. Stored at chat-rate-limit. Used solely to enforce the per-IP rate limit (5 requests per minute).
• Per-turn chat log. One blob per chat turn with the IP address, the question text, the response text, the retrieved corpus chunks, the timing, the tool uses, and the stop reason. Stored at chat-logs under a key prefixed by the UTC date. Used to debug retrieval quality and to iterate on the prompt modules.

Neither store is exposed to the public web. The function logs are accessible only to the site operator via the Netlify dashboard. The site has no concept of user accounts and does not link queries across sessions.

The chat completion itself runs against Anthropic’s Messages API. Anthropic’s data-handling policy governs the request en route to and from their API. The site does not opt into any training data sharing.

Embeddings are computed by Voyage AI; retrieval queries the Pinecone serverless index. Both providers receive the query text at request time so they can return the relevant vectors and chunks; neither receives user identifiers from this site.

Primary-source texts retrieved from Project Gutenberg and other public-domain archives retain their original status. Where a translation has been shared by its translator under specific terms (e.g., Jasper Hopkins’ Cusanus translations), those terms govern that text. See the per-philosopher provenance README in corpus/<slug>/README.md for details.

Contact

Issues and pull requests are welcome at github.com/aigamma/worldthought.com.