Discord privacy

hikomi Discord · Privacy Notice

Last updated 6 October 2026. When the Discord bot changes what it reads, sends or stores, this page changes first.

Dokumen ini juga ada dalam Bahasa Melayu: gethikomi.com/discord-privasi

The short version

The bot in the hikomi Discord server is an AI character called Hikomi, not a person. It answers when you mention it, reply to it, or write in its talk channel, and it applies a few plain moderation rules.
It stores very little: the server's settings, who pressed Accept on the rules and when, a mod-log of moderation actions kept for 90 days, and counts. It never stores the text of what anyone writes.
To write a reply, your message and the last few messages in that channel go to Anthropic. Each message meant for the bot (every message in its talk channel) also gets an automated safety check by TypeSafe. Everything the bot keeps is on our servers in Malaysia, except what Discord itself holds.
The bot never bans anyone and never sends direct messages. People, not the bot, decide anything beyond the automatic rules below.
To see or delete what the bot holds about you, email support@gethikomi.com.

What this page covers

This page covers only the hikomi Discord bot, in the hikomi Discord server. The hikomi app and websites (hikomi.ai and gethikomi.com) are covered by the privacy policy at gethikomi.com/privacy. Discord itself, as the platform, has its own privacy policy (discord.com/privacy), which applies to everything you do in Discord.

Who is responsible for your data

The bot is run by HIKOMI LABS, a business registered in Malaysia (SSM registration no. 202603256310 (TR0350156-T)), of 46, Jalan 17/43, 46400 Petaling Jaya, Selangor, Malaysia, which decides how the bot uses data and is responsible for it. For anything about your data, email support@gethikomi.com, or use the contact form at gethikomi.com/contact and pick "My data or privacy".

What the bot reads

The bot can see the messages in the channels it can access. It checks each message from members against the plain moderation rules (see "Automatic moderation") and does nothing else with a message unless that message is meant for it: it mentions the bot, replies to one of its messages, or is in the talk channel the mods chose (or a thread in it). It ignores direct messages and never sends them.

Only the messages meant for the bot, and the recent messages around them (see below), are sent to Anthropic and TypeSafe. At the moment ordinary messages elsewhere in the server are not sent to anyone for checking. A second-opinion check on ordinary messages exists in the bot's code but is switched off. If it is ever switched on, this page will say so first, and will name what is sent.

What the bot stores, and for how long

The bot keeps a small database on our server. It holds:

Server settings. Which channels and roles the bot uses, which roles members may pick for themselves, and the list of blocked words the mods set. Kept until the mods change them.
Rules acceptance. When you press Accept on the rules, your Discord user ID and the time. This is what gives you the Member role. Kept until you ask us to delete it. Leaving the server does not remove it by itself.
Role picks. The roles you choose with /roles are Discord roles on your Discord account, held by Discord. The bot's database holds only which roles are open to pick, not who picked them.
Mod-log rows. When an automatic rule fires, or a mod uses /warn, /timeout or /kick, or a member opens an appeal with /appeal: your user ID, the rule or action, the mod's user ID (empty for automatic rules), a short reason and the time. For a mod's action the reason is whatever the mod typed. Never the text of your message. Deleted automatically after 90 days. Leaving the server does not remove these rows by itself; they expire after 90 days.
Crisis counts. How many times the crisis card (see "Safety") was shown in the server on each day. A number only: no user ID, no message, no channel.
Chat spend. For each day, what the chat's AI calls cost and how many tokens and replies that was. Numbers only: no user ID, no text.
Once-only flags. Notes such as "the 80 percent budget notice for this month was sent". No member data.
Announcement drafts. Drafts of news posts about hikomi (new releases, new pages, or text the staff write), with the user ID of the staff member who posted or skipped each one and when. Which releases and pages the bot has already seen. No member data. Kept so the same news isn't drafted twice.

Some of the bot's actions are also written to places that are Discord's rather than the bot's database, and stay until the mods delete them or Discord does:

The mod-log channel, which only staff can see. Each automatic action and each mod command is posted there as a message naming the member (their mention and user ID), the channel, the rule or reason and, where there is one, a link to the message. The bot also posts notes about itself there (what the chat costs, and whether the safety check can be reached). These are ordinary Discord messages: they stay until the mods delete them, which they can do at any time. The bot does not delete them.
The under-16 note, also posted in the mod-log channel. When a message the bot answers says that the person writing it is under 16, the bot posts one note, at most one per member per day, with the member's mention and user ID, the channel and a link to the message. It never copies the message text. Nothing is done automatically; the mods decide. Like the rest of the mod-log channel, these notes stay until the mods delete them.
Appeal threads. /appeal opens a thread in the appeals channel that you and the mods can see. What you and the mods write there is yours and theirs to keep or delete as Discord allows.
Discord's audit log. When the bot or a mod times out or kicks a member, Discord records it with the reason given. Discord keeps that log under its own rules.

The bot's own log files record errors and statuses and are written not to include what people write. They are kept with our other server logs.

What the bot never stores

The text of what anyone writes. It is not written to the database, to a file or to a log.

To write a reply, the bot keeps the last 12 messages of each channel it takes part in, in memory only: each is trimmed to 500 characters, with the sender's Discord display name (not their user ID), and with the bot's own replies. In the talk channel that is every message, whether or not the bot answers it. In other channels it is only the messages meant for the bot. These are never written to disk, and they are gone when the bot restarts. If you are in a channel where the bot takes part, your display name and your messages can be among them, and can be sent along as context (see the next section).

The repeated-message rule keeps a short fingerprint of each message in memory for about 30 seconds, never the text. Rate limits (how many replies each member and each channel may have) are also held in memory only.

Who else receives your data

Discord is the platform. It holds your messages, your account and everything else you do in the server, under its own privacy policy: discord.com/privacy.
Anthropic (United States) receives the messages the bot answers, and the recent messages in that channel (up to 12, with display names, as described above), together with the character's instructions, so that Claude can write the reply. Under its commercial terms, Anthropic may not use this to train its models.
TypeSafe, whose AI model (Jev) runs the safety checks, receives each message meant for the bot (every message in its talk channel, even one the bot doesn't answer), with the recent messages in that channel before it, so that it can check for signs that someone is in danger, for someone saying they are under 16, and for a message pushing the character towards romance. It also receives the bot's reply, with the message it answers, to check that the reply doesn't promise something a bot cannot do (like remembering you later, seeing your screen, or being a person) and isn't romantic or explicit. TypeSafe processes messages outside Malaysia. We send them there only so that the bot can reply and so that the safety checks can run.
Microsoft (Azure) hosts the bot and its database, on our servers in Malaysia (Malaysia West).

That is the whole list. We don't sell or rent personal data and we don't share it for advertising. We would disclose data only if the law required it, and only what it required.

Anthropic and TypeSafe process messages outside Malaysia. Your messages go to them only because you wrote to the bot, or wrote in the channel where it listens. If you don't want that, don't write in the talk channel and don't mention or reply to the bot.

Automatic moderation

The bot applies four plain rules, with no AI: an invite link to another Discord server, five or more mentions in one message, the same message three times within 30 seconds, and a word on the blocked-words list. When one applies, the bot deletes the message, times the member out for 10 minutes, writes a mod-log row and posts a note in the mod-log channel (see above). It does not message the member.

The bot never bans anyone. It never kicks, removes roles or sends direct messages by itself. Mods, bots and anyone ranked at or above the bot are not acted on by the rules. Everything else is for the mods to decide, with /warn, /timeout and /kick, which they use themselves. If you think a decision was wrong, use /appeal in the server: it opens a thread with the mods.

Safety

Hikomi in this server is an AI character, not a person. The bot says so at the top of each thread it takes part in and once a day in the channels where it talks.

If the safety check thinks that someone may be thinking about suicide or self-harm, or may be in danger, the bot posts a fixed card as a reply in the channel, with emergency numbers and where to find help, instead of a chat reply. The card is posted as a normal message and so everyone in that channel can see it. All the bot keeps is a count for the day. If TypeSafe cannot be reached, a list of phrases is used instead.

The server is for people aged 16 and over, and 18 and over in the United States and anywhere else whose law restricts AI companions for minors. The bot does not ask for your age. If a message the bot answers says the writer is under 16, the character tells them kindly that the server is for 16 and up, and the mods get the under-16 note described above. In this server the character always talks as it would to a 16 or 17 year old: friendly, never romantic, never explicit, whatever age anyone gives. The under-16 notes are automated guesses and can be wrong.

Your rights, and how to ask

You can ask us at any time to:

tell you what the bot holds about you: your rules acceptance and any mod-log rows with your user ID;
delete those records;
correct them, or object to how we use them.

To ask, email support@gethikomi.com, or use the contact form at gethikomi.com/contact and pick "My data or privacy". Please give your Discord user ID (in Discord, turn on Developer Mode in Advanced settings, then right-click your name and pick Copy User ID), so that we can find the rows. We answer privacy requests within 21 days. If you're not happy with our answer, you can complain to the Personal Data Protection Department of Malaysia (Jabatan Perlindungan Data Peribadi), or to the data protection authority where you live.

Leaving the server does not by itself delete anything the bot holds about you: mod-log rows expire after 90 days, and a rules acceptance stays until you ask for it to be deleted. The notes in the mod-log channel are Discord messages: when you ask, we ask the mods to delete the ones about you. Your own messages, your Discord account and Discord's audit log are Discord's: ask Discord about those.

The moderation rules look at every member's message in the server, but they keep none of them.

Changes to this page

If we change this page in a way that matters, we tell you in the server before the change applies. The date at the top shows when it last changed.