soft questions, soft answers ✦

yes — every command, every dashboard page, every module. there is no premium tier, no paywalled feature, no ad. hosting is covered by @volc and inference runs on free-tier providers with automatic failover. if she saves you time, a coffee is always appreciated but never required.

the minimum for each feature to work: conversation history (last 20 messages per user per channel) for ai chat, a capped list of durable facts for memory, xp counters for leveling, warnings for moderation, and feature settings per server. no analytics, no third-party trackers, no selling anything. see the privacy module guide for the full controls.

both, built in: /privacy export gives you everything as json, /privacy delete erases it everywhere after a double confirmation. per-feature opt-outs are one command away too.

aurelia is currently hosted and maintained as a managed project — the source repository is private. that is separate from using her: anyone can add the bot to their server for free (see the getting-started guide), and she runs as one shared, carefully tended instance. the support server is the place to ask if you need something self-host-shaped.

several, behind one router. everyday chat starts on gemini flash; mistral small answers as fallback and handles sensitive work (moderation, automod) by default; a glm reasoning model takes the genuinely hard questions under a daily budget; and groq models are the emergency floor that keeps her talking when everything else is rate-limited. whichever answers, responses are post-processed so reasoning never leaks into chat — only the answer does. see the ai privacy notes below.

aurelia can use multiple external ai inference providers, and which one serves a request depends on the request type and live provider availability. the rules: only the minimum context a feature needs is sent; moderation and automod traffic is routed to providers chosen for stricter data terms by default (never the free-tier consumer endpoints unless explicitly configured); your /privacy opt-outs are enforced before any provider is invoked; provider telemetry stores counts and latencies, never conversations; and api keys stay server-side — none ever reach the browser.

the support server (link coming soon) is fastest. bug reports and feature requests both land there — they go straight into the build queue.

yes — settings, xp, warnings and ai memory are all per-server, and the dashboard lists every server where you hold the manage server permission. reminders follow you across servers because they are per-user.

reminders and recurring tasks catch up on restart by design — nothing fires twice, nothing is lost. the public stats page shows live vitals and uptime history.