1. What data this tool handles
When you use this tool, the following is sent to the server so it can respond: the agent name you type, the traits you choose and their intensity, and any prompt text you enter to generate a response.
2. How it's stored
All of this is kept only in the server's in-memory process state and is not retained after the server process restarts or stops. The one exception is the main mode's aggregate voting data - see section 8 below for exactly what that includes.
3. No third-party sharing
This tool does not send your data to any third-party service, analytics provider, or advertising network. There are no tracking scripts, cookies, or external calls made on your behalf.
4. No accounts or personal identifiers
There is no login system, and the tool does not knowingly collect names, email addresses, or other personal identifiers beyond whatever text you choose to type into the agent-name or prompt fields.
5. Server logs
Like most web servers, the underlying framework may write basic request logs (e.g., timestamps, paths, status codes) to its own console/log output for debugging. These logs are local to whoever is running the server and are not transmitted elsewhere by this application.
6. Your choices
Because nothing is persisted, you can clear your data at any time simply by not creating new agents, or by restarting the server, which discards all in-memory state.
7. Changes
If how this tool handles data changes in the future (for example, if persistent storage is added), this policy will be updated to reflect that.
8. Main analytics data
Stored, in a small server-side database: the timestamp of each Main round, the 1-3 trait names randomly assigned to each of the two options, each reply's length in characters, and which option (if any) you voted for.
Not stored: the prompt text you typed, the reply text itself, your IP address, or any session or account identifier. There is no way to trace a stored round back to the person who triggered it.
This aggregated, non-identifying data is shown publicly on the Analytics page and updates for everyone in real time. Because it lives on disk rather than only in memory, it can survive a server restart - though depending on how this tool is hosted, it may still be cleared when a new version is deployed.
9. Raw/no-trait comparison data
On a randomly selected minority of rounds, one option is generated with no personality traits at all instead of the usual random 1-3, so its side of that round is stored the same way as any other option, just with an empty trait list. Additionally stored for these specific rounds: which of the two options (A or B) was the trait-free one. No new categories of data beyond what's listed in section 8 are collected - the same non-identifying, structural-only fields apply, and the aggregate result (how often the trait-bearing side won) is shown publicly on the Analytics page.