This policy explains what kesarAI processes when you add it to a Discord server or use one of its commands — and what we do not do with that information.
Who We Are
kesarAI (the “Bot”) is operated as an individual project. The operator acts as the data controller for the purposes of the EU General Data Protection Regulation (GDPR) and the Swiss Federal Act on Data Protection (FADP).
Contact the operator through the kesarAI support server for privacy questions, data-subject requests, and security reports.
What We Collect
kesarAI processes only the information required to run the feature you invoke.
Collected automatically
- Guild metadata — numeric guild, channel, and role IDs (and their names) so configuration can be displayed and applied.
- User identifiers — numeric Discord user IDs for permission checks, the per-user AI battery and rate limits, and audit-trail attribution.
- Command invocations — the slash command used, the channel it was used in, and the timestamp.
When you use AI chat
- Your messages to kesarAI — messages that start with
ksr, @mention the Bot, reply to it, or use/ai ask, plus any attachments. - Recent channel messages (room view) — kept in memory and read only when a question needs the recent chat; not stored by kesarAI (see How AI Chat Works below).
- Your conversation — kept for 30 minutes after your last message.
- Full AI conversation logs (temporary) — while kesarAI moves to Claude, each AI answer is logged in full (your message, the context sent to the AI including recent channel messages, the reply, tools used, timings and cost) so answers can be improved and waste found. Logs are deleted automatically after 14 days, and
/privacy forgetdeletes yours straight away. People who paused AI memory in/privacyare not logged. - Memory facts — facts about you that kesarAI learns from your public messages to it, and facts you ask it to save.
- Reply links — which kesarAI reply answered whom (message IDs only, 7 days).
- Usage records — model, token counts, estimated cost, timings and which tools ran, per AI call, with no message text.
- Your AI battery — your battery level, any boost or one-time credit, today’s image count, and a log of charges and denials — no message text.
Collected only when relevant
- Message content is read on demand for features such as
/ai summarise,/ai fallacy, and/ai modadvice. It is forwarded to the AI backend for that single request and is not retained afterwards, except where audit logging is explicitly enabled. - Audit events — moderation actions, member events, and AI usage are written to
audit.dbwhen the corresponding category is enabled for your guild. - Roles when you leave a server — the role IDs you held and your Discord username, so they can be given back if you rejoin (see Sticky Roles below).
- Booster roles — if you boost a server that uses them, the design of the role you made and your requests to change it (see Booster Roles below).
- Levels — in servers that use leveling, the XP and level you have earned there.
- Starboard — if you star a message in a server that uses the starboard, the fact that you starred it (IDs only, no message text).
- Voice rooms — if you open a room from a join-to-create channel, the room’s name and settings and who you let in or removed.
- Giveaways — if you enter a giveaway, that you entered and whether you won.
How AI Chat Works
Who answers
Chats with kesarAI are answered by default by Google’s Gemini 3.5 Flash-Lite through Google Vertex AI. If you choose a different model on your account page, your messages are answered by that model instead: Anthropic’s Claude Haiku 5.5 through Anthropic’s API, the others through Google Vertex AI. ksr think and /ai think use the model one step up from the one you chat with (Gemini 3.8 Flash, or Claude Sonnet 5.5 if you chat on Claude Haiku) unless you pick a separate think model there. Whichever model you use, long conversations are shortened by Claude Haiku, so a summary request containing your conversation is sent to Anthropic. If the bot owner turns on low power (an optional setting, off by default), some features pause while your battery is low; it never changes which model answers you.
What the model receives
Your message and attachments, your recent conversation with kesarAI in that channel, memory facts and preferences that apply to you, server details kesarAI looks up to answer (such as channel, member, role and level information) and, on Claude, the room view below.
Room view
kesarAI keeps the last ~30 minutes of a channel (up to about 40 lines) in memory and reads them only when a question needs them, such as “who’s right?” or “summarise the chat”. Those lines are never stored by kesarAI and never used for AI memory; they go to the AI provider (Anthropic) only when they are read to answer a question. kesarAI’s reply may mention what it saw. Like any reply, it stays in the asker’s conversation for up to 30 minutes, and it is removed when anyone whose messages it saw forgets their AI data, pauses AI memory or clears their conversation. Room view skips ticket and verify channels, (private) asks and anyone who paused AI memory, and deleted messages drop out immediately. Room view is used only with Claude, not with other models. Server administrators can turn room view off for their server (/setup → AI Access → Room View).
Web search
When an answer needs current information, kesarAI searches the web. On Gemini (the default) the search runs through Google Search via Gemini; on Claude it uses the Brave Search API, and if Brave is unavailable or out of credit, the same search runs through Anthropic’s web search tool instead. The search provider receives only the search words the AI writes, sent from kesarAI’s server — not your Discord ID, username or IP address. Before a search is sent, kesarAI removes Discord mentions, IDs and @handles from it, and the AI is instructed not to put members’ names or personal details into searches; a search can still paraphrase what you asked. With a chosen model that can’t use kesarAI’s search, a search may also run through Google Search via Gemini. Search results are held only in memory while the answer is written; they are not stored in kesarAI’s memory, logs or analytics (usage records note only that a search happened, its cost and timing). The answer, with up to three source links, is posted in Discord and kept in your conversation like any other reply. The Truth Check right-click command and ksr fact check search the same way.
Links and images
On Claude, links you ask kesarAI to read are fetched by Anthropic’s web fetch tool or, depending on configuration, by kesarAI itself; Discord links are never fetched. Images in the conversation are sent to Anthropic. On other models kesarAI fetches links itself, and images go to the model you chose. Voice messages are transcribed by Google Gemini; YouTube captions are fetched by kesarAI directly. Messages from people who paused AI memory are never read, viewed or transcribed for other people.
Memory
kesarAI learns facts about you from your public messages to it (for example your interests or what you’re working on) and keeps them private to you. It also saves a fact when you ask it to remember something; those are private unless you say “publicly” (visible in that server) or “everywhere” (every server you share with kesarAI), and the confirmation tells you who can see it. Facts you asked it to save stay until you delete them; facts it learned on its own expire within 180 days (7 days for short-lived activity). Deleting your own message removes the facts learned from it; a server’s bulk cleanup of a channel does not. You can pause memory at any time (/privacy, or Privacy on your account page); while paused nothing new is saved, your messages are left out of room view, and kesarAI will not quote your messages to the AI when someone else replies to them.
Conversations
You have a separate conversation with kesarAI in each channel. It resets after 30 minutes without a message and is never picked up again after that; its stored copy is removed within the following half hour (a cleanup runs every 30 minutes), and private conversations are kept in memory only. Long conversations are summarised by Anthropic’s Claude Haiku to keep context short, for every chat model you can pick (Google Gemini Flash-Lite is used instead only if the Anthropic API is not configured); a summary lives only inside that conversation and expires with it. Deleting a Discord message removes it from room view at once, but if it was already part of an open kesarAI conversation, it stays in that conversation until the conversation resets. ksr reset clears your conversation in the current channel; ksr forget and /ai forget clear all of them. When you forget or pause, your own conversations are deleted and anything you said that appears in other people’s open conversations is removed from them. To let replies continue a conversation, kesarAI keeps a 7-day record linking its replies to the person who asked (message IDs only, no text). A few short-lived settings, such as “stop joking” for 30 minutes, are kept in memory only.
Your AI battery
To run the per-user AI battery, cost limits and abuse prevention, kesarAI records for each AI call your user ID, server ID, the command, the model, token counts, the estimated cost, timings and which tools ran — never message text. These records are deleted after 180 days. kesarAI also stores your battery level, when it last changed, any boost or one-time credit the bot owner gave you, and today’s image count; a full battery that has been idle for 30 days is deleted, unless a boost or credit is on record. Each battery denial and command charge is logged (time, command, cost and battery level) and kept for 30 days. Deleting your AI data (/privacy forget, or the Privacy page on your account) removes your memory, preferences and conversations, but not these usage records or your battery level, which are kept for fair-use accounting.
Testing answer quality
To test the AI’s quality we use rewritten, anonymised examples based on requests people have made; pausing AI memory removes yours from future test sets.
Why We Process It (Legal Basis)
Under GDPR Article 6 and the equivalent provisions of the Swiss FADP, we rely on the following legal bases:
- Legitimate interest — running moderation tooling, preventing abuse of the AI surface, and enforcing the per-user AI battery and per-user and per-guild rate limits; reading the recent conversation in a channel when a question there needs it (room view), limited to the last ~30 minutes, kept in memory only, and avoidable by pausing AI memory; and keeping usage and battery records to run the AI battery and prevent abuse.
- Performance of the service — storing the configuration you create, routing your commands, and returning their results.
- Consent — sending your message, attachments and context to the AI provider when you talk to kesarAI or invoke an AI command. You can decline by not invoking it.
Data Storage & Retention
Persistent data lives in SQLite databases on the operator’s hosting infrastructure:
config.db— guild configuration (channel mappings, role grants, feature toggles) and your account preferences (verbosity, chat model, personality, voice). Kept while the Bot is in your server.audit.db— audit-trail events, subject to the retention policy set per guild.ai_memory.db— facts you asked kesarAI to save (until you delete them); facts it learned from your messages (180 days; 7 for activity); copies of public conversations (removed within an hour of the last message; private ones are memory-only); reply links (7 days, IDs only).ai_transcripts.db— temporary full AI conversation logs (14 days, then deleted automatically; not kept for people who paused AI memory).ai_usage.db— per-call cost and usage records (180 days); the battery denial and charge log (30 days); your battery level (deleted after 30 idle days once full, unless a boost or credit is on record). No message text.- Privacy actions — when you pause memory or delete AI data, kesarAI records that action (IDs and one-way fingerprints only, no message text) in a separate privacy log, so your deletions and pauses stay in force even if an older backup is restored. This log is kept for as long as kesarAI keeps AI memory.
sticky_roles.db— the roles members held when they left a server (see Sticky Roles below).booster_roles.db— boosters’ role designs, change requests and linked accounts, in servers that use booster roles (see Booster Roles below).leveling.db— the XP and level each member has earned, in servers that use leveling. A server can switch on a public leaderboard link; anyone who has that link sees the display names, avatars and levels on it.starboard.db— in servers that use the starboard, which messages reached a board (message, channel and author IDs and the star count; no message text) and which members starred them. Removed when the board is deleted.voice_rooms.db— in servers that use voice rooms, the rooms members opened (owner, name, limit, locked or hidden), who was let in or removed, and each member’s saved room settings. Closed rooms are deleted after 90 days.giveaways.db— the giveaways a server runs, who entered and who won. Entry lists are deleted 30 days after a giveaway ends; winners stay with the giveaway.- In servers where message indexing is switched on, an index of channel messages, kept for up to 30 days.
Never stored by kesarAI: room-view lines, web search results, and page text fetched for you. Nightly backups of these databases are kept for 14 days, so deleted data can remain in a backup for up to 14 days.
Audit logs are retained indefinitely by default. Server administrators can configure a per-guild retention period using /logs retention <days>, after which older entries are removed by a daily purge task. Configuration data persists for as long as the Bot is in your guild, plus a short grace period after removal.
Sticky Roles
When you leave a server that uses kesarAI:
- Where sticky roles is on, it saves the IDs of the roles you held, your Discord username, when and how you left (left, kicked or banned), and whether you were jailed or quarantined there.
- Where sticky roles is off, it saves nothing, unless you were jailed or quarantined (then only that marker and when you left), or an old quarantine of yours is still on record in that server’s moderation log (then only a note that you left without one, so it isn’t put back by mistake). No roles or username are saved.
To tell whether you were quarantined when you left, kesarAI also reads that server’s existing moderation log (quarantines, verifications, and the roles you held when you left, where the server logs member activity). If you rejoin, this is used once:
- Where the server has sticky roles turned on, your roles are given back, within the server’s rules. Sticky roles is off by default, unless kesarAI gives that server its join auto-roles.
- A jail or quarantine is put back in every server, even with sticky roles off, so leaving can’t be used to escape one.
Retention. A saved role list is kept for 7 days after you leave by default; each server can choose a different period, up to forever. Once it has been used on your rejoin it is kept 7 more days, so moderators can see what was given back, then deleted. A daily task deletes expired records. A jail or quarantine marker doesn’t expire: it stays until you rejoin and the jail or quarantine is put back, and a moderator lifts it from there. A note that you left without a punishment is kept for up to a year.
/privacy forget and “Delete all my data” on your account’s Privacy page delete your saved role lists and usernames, and any notes that you left without a punishment, in every server. A jail or quarantine marker stays, as a moderation record.
Booster Roles
Where a server uses booster roles, a member who boosts it can make a role of their own with /boosterrole. kesarAI then stores, for that server:
- your role’s name, colours and icon (an image you uploaded or an emoji), and the role’s Discord ID;
- your change requests, and what staff answered;
- the accounts staff linked to your role, if any;
- when your boost started and ended, when your role is due to be removed, and when you can next change it;
- the roles kesarAI gave you for boosting or for wearing the tag, and why, so it takes back only those;
- whether you wear the server’s tag, only in a server that gives a role for it.
If staff link your account to someone else’s booster role, kesarAI stores that link, with who made it and when, until it is removed or the role is gone. Role changes are also written to that server’s audit log under Members, with who asked and who approved. A saved design is kept so your role comes back if you boost again.
Removal. /privacy forget and “Delete all my data” on your account’s Privacy page delete your saved designs, your change requests and your server-tag state, in every server. Your live role stays until your boost ends, and kesarAI still removes it then, so the record of the role, its due times and any linked accounts stays until the role is gone. You can remove the role sooner with Remove my role in /boosterrole. If your tag changes in a server that gives a role for it, kesarAI notes it again.
Third-Party Services
Discord API
All bot activity flows through the official Discord API. Discord’s own Privacy Policy governs how Discord handles message data, identifiers, and server membership.
Anthropic (Claude)
Anthropic’s Claude models, called through Anthropic’s first-party API, answer chats for people who pick Claude Haiku 5.5, and think turns on Claude Sonnet 5.5 (the default think model for Claude Haiku users, or a pick). They also handle Truth Check (and ksr fact check when you chat on Claude), summaries of long conversations for every chat model, the backup web search, and reading links and images on Claude. A chat prompt may include your message and any attachments you share, your recent conversation, the room view, server context the Bot reads to answer (such as channel, role, and member details), and memory facts and preferences that apply to you. Anthropic processes this content as our service provider to generate the reply.
Under Anthropic’s Commercial Terms of Service, Anthropic may not train its models on content submitted through the API, and that content is processed under Anthropic’s Data Processing Addendum (DPA). Per Anthropic’s published retention policy, API inputs and outputs are deleted automatically within 30 days, except where Anthropic must keep them longer to enforce its Usage Policy (content flagged as a violation may be kept for up to two years) or to comply with the law.
Brave Search
Web searches on Claude go to the Brave Search API, operated by Brave Software, Inc. Brave receives the search words only. How Brave processes API requests is governed by its Search API Terms of Service.
Google Cloud
Google Cloud runs the default chat model (Gemini 3.5 Flash-Lite, with web searches through Google Search) and the other non-Claude models, image generation and editing (including the Edit Image right-click command), transcription of voice messages and of audio and video attachments (ktr), automated moderation (autopilot, together with Cloud Vision and Web Risk, and a periodic safety scan of /say messages), auto-translate, ModAI, the Setup AI assistant, the Summarize right-click command, and the /ai summarise, /ai fallacy, /ai modadvice and /ai draftwarn commands. Text-to-speech uses Google Cloud Text-to-Speech. These features send prompts, which may include message content scoped to the request, to Google Vertex AI, the enterprise Gemini surface on Google Cloud. We deliberately use Vertex rather than the consumer Gemini API because of its stronger data-handling guarantees: per Google’s Vertex AI data governance policy, customer prompts and responses are not used to train, fine-tune, or otherwise develop Google’s foundation models, and they are processed under the Cloud Data Processing Addendum (CDPA) as “customer data,” not as input to Google’s own services.
Even with these protections, do not share through AI commands anything you would not be comfortable sending to a third-party service provider.
International Transfers
Vertex AI processes prompts on Google Cloud infrastructure that may be located outside Switzerland and the European Economic Area. Transfers to such locations are governed by Google’s Cloud Data Processing Addendum and the Standard Contractual Clauses incorporated by it.
Anthropic processes API data in the United States and other countries where it or its subprocessors operate; transfers are governed by Anthropic’s Data Processing Addendum, which incorporates the EU Standard Contractual Clauses together with UK and Swiss addenda.
Brave Software, Inc. is based in the United States; how it processes API requests is governed by its Search API Terms of Service.
By using AI commands you accept that the content of those prompts is transferred under these frameworks.
Data Security
Traffic between the Bot and Discord, Anthropic, Brave, and Google is encrypted in transit using TLS. The SQLite databases live on the operator’s hosting infrastructure with restricted filesystem access; no Discord, Anthropic, Brave, or Google credentials are written to the audit trail. No security measure is perfect; in the event of a breach that materially affects your data we will notify affected administrators through the support server.
Your Rights
Subject to applicable law, you have the right to access, correct, export, delete, or restrict the processing of personal data we hold about you, and to object to processing or withdraw consent.
You can export your AI data, delete your memory, preferences, conversations, saved sticky-role lists, saved booster role designs, starboard stars, voice room settings and entries in finished giveaways, and pause memory yourself with /privacy in Discord or on your account’s Privacy page. Usage and battery records are not removed by these tools; they are kept for the periods above. To ask about anything you can’t delete yourself, contact us in the support server.
To exercise these rights, contact us through the support server with enough information to identify the relevant Discord user ID and guild ID. Server administrators can clear audit logs and configuration for their own guild using the Bot’s built-in controls; end users who wish to be removed from a guild’s audit history should also contact that guild’s administrators.
You have the right to lodge a complaint with a supervisory authority. Residents of Switzerland may contact the Federal Data Protection and Information Commissioner (FDPIC); residents of the EU/EEA may contact their local data-protection authority.
No Sale of Data
We do not sell, rent, or trade personal data. We do not use it to train external AI models, and we do not build cross-server profiles of individual Discord users. Web search results are not stored and are not used to train any model.
Children’s Privacy
The Bot operates on Discord, which requires users to be at least 13 years old (or the higher minimum age specified in their jurisdiction). We do not knowingly collect data from users below that age. If you believe a minor has interacted with the Bot in a way that requires intervention, contact us so we can act.
Changes to This Policy
We may revise this Privacy Policy from time to time. When we do, the “Last updated” date at the top of this page changes, and material revisions are announced through the support server.
Contact
For privacy questions, data-subject requests, and security reports, reach out via the kesarAI support server.
