Website operation · 16 November 2026
Understand what leaves your browser.
Guided study and local sandbox
The guided study uses fully synthetic records and generator labels. Calculations happen in your browser. The optional sandbox parses uploaded files locally and uses a teaching mock; it does not send those records to the hosted model. Downloaded research records may contain the files you loaded. Keep them within your research environment.
Anonymous live examples
Live examples accept fixed example identifiers only. The Worker selects synthetic text or topic words on the server and sends them to OpenAI using GPT-4o. You cannot upload a research corpus or submit an arbitrary prompt to this public endpoint. Provider credentials stay in Cloudflare server secrets.
A signed HttpOnly, Secure, SameSite=Strict session cookie lasts one day and supports a ten-call daily allowance. A daily keyed hash of the connection's IP address supports a thirty-call shared-network limit. The application does not store raw IP addresses or email addresses. Session/IP counters are used for abuse and spending controls, not advertising.
Short model-response receipts support duplicate-request handling and expire after thirty minutes, with scheduled cleanup. Aggregate usage counters contain call counts and conservatively estimated spending. These are separate from Cloudflare's infrastructure processing and recovery systems. OpenAI receives the selected synthetic inputs; calls use store:false, which does not mean zero provider retention.
Service limits
The public live demo has a site-wide OpenAI allowance of $1 per UTC day and $5 per UTC calendar month, plus concurrency and request limits. Reservations are persisted before requests are sent. An uncertain request retains its allowance; completed calls record returned usage. These limits cover estimated OpenAI token charges at the documented model price, not Cloudflare infrastructure charges.
When live access is paused, busy, unavailable or out of allowance, the guided walkthrough, documentation and downloads remain available. Anonymous session and IP limits are abuse controls; they do not identify a person, and shared networks may share an allowance.
Research data belong in your environment
For your own records, fork the repository or install the Python package. Use the API account, computing environment, access arrangements and data-processing authorization approved for your project. TopicLLM's synthetic demonstration does not establish permission to send real research data to a provider.
Start the local workflow · Report a software issue without private data