DocsReference
Does Astri use AI?
Yes, for decorative parts like page text and embeds; people build everything that protects your server.
Short answer
Yes. Astri uses AI to help write its decorative parts, like the text on these pages and the wording of the bot's embeds. Everything that protects your server, from the backend to the rules that decide when Astri acts, is built by people. A person reviews all AI-written code before it's published, and the bot never uses AI to make a decision.
Decorative and critical
Astri splits its work by what happens if something goes wrong. A mistake in a decorative part means a sentence reads badly or an embed looks off. A mistake in a critical part could ban the wrong member, miss a real attack or expose your server's data.
| Part | Kind | Built by |
|---|---|---|
| Text on the website and in the docs | Decorative | AI helps, a person reviews |
| Wording and look of the bot's embeds | Decorative | AI helps, a person reviews |
| Detection rules and thresholds | Critical | People |
| Bans, role strips, lockdown, quarantine and undo | Critical | People |
| Rollback and backups | Critical | People |
| Database and data handling | Critical | People |
| Sign-in, permissions and the API | Critical | People |
What AI helps write
Page text
AI helps draft the words on the frontend: the website and these docs. A person checks each page against what the bot really does, so the numbers and behaviour you read here match it.
Embeds
AI helps with how the bot's embeds look and read: alerts, incident reports and command replies. It shapes how a message is worded, never what triggers it or what Astri does next.
What people build
The backend
The backend is the bot itself. It reads your server's audit log and events through Discord's API, keeps score and acts. It holds Administrator in your server, so people write it.
Detection
The anti-nuke budget, raid risk scores, the honeypot and anti-spam are hand-written rules with fixed numbers. Each docs page lists them, so you can see exactly when Astri will act.
Actions and undo
Anything that changes your server is written by people: bans, kicks, timeouts, stripping roles, lockdown, quarantine, rollback and undo.
Data and access
The database and the code that stores, reads and deletes your server's data are written by people. So are sign-in with Discord and the checks on who can see and change what.
How AI-written code is checked
- AI suggests a change, such as new page text or a reworded embed.
- A person reads all of it in a code review, and rewrites or rejects anything that isn't right.
- Only then is it published.
Nothing AI writes goes live without a person approving it. The bot also has automated tests for detection, actions and rollback, which catch a change that breaks how Astri behaves.
Does the bot use AI to decide?
No. There's no AI model inside Astri. Every decision is rules-based: a budget running out, a risk score crossing a level, a message in the honeypot. The same event always gets the same result.
That matters for a security bot:
- Predictable: you can read the rules and know what Astri will do.
- Explainable: each incident shows what Astri saw and what it did about it.
- Hard to trick: there's no prompt an attacker can type to talk Astri out of acting.
Your server's data and AI
Astri never sends your server's messages, members or settings to an AI model, so none of it can be used to train one. Privacy has what Astri stores and for how long.
Checking for yourself
Shadow mode lets you watch every call Astri would make before it acts on any of them, so you can compare its decisions with the rules in these docs.