Frequently asked, honestly answered
Answers grounded in how Sloth Lee actually behaves — every claim here is observable in the product.
Can the AI actually change my server?
Yes — through tools with the same permission gates as slash commands. Every write needs a Confirm click from staff before anything happens, and every action is audit-logged with who asked, what tool, and the result. Destructive actions are never silent.
What happens if nobody clicks Confirm?
Nothing. Proposals expire on their own. The bot never auto-applies a change the model proposed, and it never resumes work on its own after a confirmation.
Do members get AI tools?
Members get read-only surfaces: knowledge search over staff-curated entries and channel topics they can already see, plus channel summaries. All operational tools live in staff mode.
Who pays for the AI?
Each server brings its own Ollama Cloud key, encrypted at rest. Daily request caps apply per server (default 200), plus separate caps for background work like digests. Staff can see exactly what was spent in /ai status or the dashboard's Budget page.
Where does my data live?
In your Postgres database. Conversations older than 30 days are excluded at read time and pruned daily. Chat logs are encrypted. Backups are encrypted and restore-tested. Self-hosted means you hold all of it; the hosted instance is the same software.
Is there a premium tier?
No. Both ways to run Sloth Lee — invited or self-hosted — are free and are the same software. See pricing.
What does the bot need permission-wise?
It asks for the Discord permissions its modules actually use. Within the server, the AI respects each staff member's own Discord permissions — it cannot act above the rank of the person who asked.
What if the process restarts mid-giveaway?
Scheduled work is durable: jobs are rows in Postgres, claimed exactly once, and missed windows replay safely on boot. A giveaway due during downtime still ends — exactly once.
Can I run two copies for redundancy?
Yes, now: scheduled sweeps are fenced by a leadership lease, so a second process idles instead of double-sending. Until you need that, one process per deployment remains the simple default.
How do I know the bot is up?
The /status page shows live health straight from the bot's own health endpoint: gateway, database, scheduler queue, and leadership. Self-hosted instances expose the same JSON at /healthz?enriched=1.
Self-hosting instead? The setup guide takes about ten minutes.