What the word filter actually sees
Axolmod checks messages against a server's filtered-word list. Message text travels to the API for that check — and is never written to the database. Here is the full path a message takes.
- Moderation
- Privacy
A word filter is only useful if it can read messages, and that is exactly the part worth being precise about. This post describes what actually happens when Axolmod is enabled in a server.
The path of a message
- A member posts a message in a server that has Axolmod's filter enabled.
- The bot sends the text of that message to the Axolix API.
- The API compares it against that server's filtered-word list and returns a yes or no.
- On a match, the bot deletes the message on the server's behalf.
The comparison happens against a list the server's own managers configured from the dashboard. The bot does not decide what to block; it applies the list the owner gave it.
The message text is not stored
This is the part that matters most. Message content is checked and discarded. It is not written to the database, not kept in a log, and not retained for review. There is no archive of what people said in your server sitting on our side.
That constraint shapes the design. Every filtered word a server configures lives in that server's own list, and the check is a single comparison against it — there is nothing to build a profile from, because nothing is kept.
Why the list is cached briefly
Every message in every server using the filter would otherwise mean one database read per message. The list is therefore cached in memory for thirty seconds per server.
The accepted trade-off: a word added or removed from the dashboard can take up to thirty seconds to take effect on incoming messages. Writes never touch the cache, so there is nothing that can drift out of sync — a cache entry is only ever produced by a query that actually succeeded.
Who is responsible
The server owner decides what the filter blocks and is responsible for that choice being lawful and fair to their members. Axolix applies the configuration; it does not review the moderation decisions themselves.