# Instance settings

The settings that define how your instance is run: registration mode, post length,
default locale, who can see the timelines, the instance name, description, terms, rules, theme, and custom emoji.

## Configuration and instance settings

| | Configuration | Instance settings |
| --- | --- | --- |
| What | How the deployment is built: roles, adapters, endpoints, host names, secrets | How the instance is run: registration mode, post length limit, default locale, who can see the timelines, the instance name, description, and terms |
| Where | Environment variables `MY_*` and the configuration file ([configuration reference](https://mycroblog.dev/next/reference/configuration/)) | The database. Every change is kept as a revision |
| Changing | Takes effect when the process restarts | Takes effect immediately with `my instance set`, the admin screens of the web client, or the client API ([admin screens and the client API](#admin-screens-and-the-client-api)) |
| When unset | The default in the schema (the server does not start if the key is required) | The default defined in the code |

- **No key is in both.** Neither overrides the other. You cannot give an instance setting through an environment variable or
  the configuration file, and you cannot store a configuration value in the database.
- A value that is fixed per deployment and changes the structure of the process is configuration; a value that changes while
  the instance is running is an instance setting.

## Prerequisites

- The configuration is given as described in the [configuration reference](https://mycroblog.dev/next/reference/configuration/) (`my config check`).
- The schema is applied to the database (`my schema apply`; see [Database](https://mycroblog.dev/next/operations/database/)).

## Keys

### Settings (one key, one value)

| Key | Type | Default | Description |
| --- | --- | --- | --- |
| `registration_mode` | `open` / `approval` / `invite` / `closed` | `approval` | The registration mode ([registration](https://mycroblog.dev/next/operations/registration/)). While a service shutdown is announced and after it is closed, and on a [one-person server](#one-person-server), the effective value is `closed`. The stored value does not change, and cancelling the announcement or turning `single_user` off restores it. `GET /instance` and NodeInfo show the effective value |
| `max_status_length` | integer 1-100000 | `500` | The maximum length of a post, in characters |
| `default_locale` | BCP 47 language tag | `en` | The default locale. It is the initial UI locale of new users, and the fallback when no catalog matches a user's preference |
| `local_timeline_visibility` | `off` / `signed_in` / `everyone` | `everyone` | Who can see the local timeline (`GET /api/v1/timelines/local`): nobody, signed-in users, or everyone. `off` answers 404 and `signed_in` answers visitors who are not signed in with 401; in both cases the navigation hides it. Unless the value is `everyone`, the landing page shows no sample posts. On a [one-person server](#one-person-server) it reads as `off` |
| `federated_timeline_visibility` | `off` / `signed_in` / `everyone` | `signed_in` | Who can see the federated timeline (`GET /api/v1/timelines/federated`). The values mean the same as for the local timeline, and it reads as `off` on a one-person server too |
| `preview_image` | media ID (empty = unset) | empty | The link preview image (`og:image`) for the pages without an image of their own, under "Link previews" on Settings → General. When unset, the thumbnail, then the logo, then the bundled image is used ([link previews](https://mycroblog.dev/next/operations/link-previews/)) |
| `contact_account` | username of a local account (empty = unset) | empty | The account to contact the operators. See [contact and thumbnail](#contact-and-thumbnail) |
| `contact_email` | email address (empty = unset) | empty | The email address to contact the operators. It is also the `Reply-To` and the signature of the emails the server sends |
| `thumbnail` | media ID (empty = unset) | empty | An image that represents the instance (thumbnail), on the about and landing pages, and in link previews when `preview_image` is unset |
| `remote_media_policy` | `none` / `avatars` / `all` | `none` | Whether to keep copies of remote media. `none`: the media proxy relays every request. `avatars`: avatars, headers, and custom emoji are cached in blob storage for `remote_media_cache_ttl_seconds`. `all`: all remote media is. The total size of the cache is limited by the configuration `remote_media_cache_max_bytes` |
| `remote_media_cache_ttl_seconds` | integer 60-31536000 (seconds) | `86400` | How long fetched remote media may be kept. It is also the `Cache-Control: max-age` of the media proxy responses, so CDNs and browsers cache for this long |
| `remote_media_cache_max_bytes` | integer from 1 (bytes) | `1073741824` | The size limit of the copies of remote media the media proxy keeps. The hourly cleanup deletes the expired copies, then the least recently accessed ones over this limit |
| `authorized_fetch` | `true` / `false` | `false` | Authorized Fetch (secure mode). When `true`, an ActivityPub GET without a verifiable HTTP signature is answered with `401` for objects and collections, and with a minimal document that carries the key for a user's actor. The instance actor always returns its whole document, so any server can fetch the key to verify a signature |
| `federation_mode` | `off` / `inbound` / `outbound` / `both` | `both` | The direction of federation. `off`: no federation. `inbound`: you follow and read remote accounts, but local posts, profiles, and reactions are not sent out, and follows from outside are refused (`Reject`). `outbound`: you accept follows from outside and deliver local posts, but remote posts and reactions are not taken in. `both`: send and receive. A change takes effect at once and needs the permission `manage_settings` |
| `federation_scope` | `open` / `allowlist` | `open` | Which servers this server federates with. `open`: every server that no domain action suspends. `allowlist`: only the domains on the allowlist and their subdomains; every other server is treated as suspended, and ActivityPub GETs need a signature from an allowed server whatever `authorized_fetch` says. Switching to `allowlist` ends the follows with the servers it cuts off for good; see [federating only with allowed servers](https://mycroblog.dev/next/operations/moderation/#federating-only-with-allowed-servers). A change takes effect at once and needs the permission `manage_settings` |
| `dynamic_client_registration` | `true` / `false` | `true` | Dynamic client registration (RFC 7591). While `true`, `POST /oauth/register` accepts the registration of public clients without authentication. When `false`, the endpoint refuses every request with `403` (`access_denied`); clients that are already registered keep working. A periodic job deletes dynamically registered clients that are not used, following the configuration `dynamic_client_unredeemed_retention` and `dynamic_client_idle_retention`. A change takes effect at once and needs `manage_settings` |
| `domain_actions_visibility` | `off` / `signed_in` / `everyone` | `off` | Who can see the list of domains with actions taken (`GET /api/v1/instance/domain_actions`): nobody, signed-in users, or everyone ([moderation](https://mycroblog.dev/next/operations/moderation/#publishing-the-domains-you-act-on)) |
| `domain_actions_rationale_visibility` | `off` / `signed_in` / `everyone` | `off` | Who can see the public comments of that list. Someone who cannot see the list never sees them |
| `appeal_window_days` | integer 20-365 (days) | `180` | How many days after an action the person concerned can appeal ([moderation](https://mycroblog.dev/next/operations/moderation/#appeals)). Changing it also changes the deadline of actions already taken |
| `session_origins_retention_days` | integer 90-3650 (days) | `180` | How many days after a login to keep the IP address and User-Agent of the login ([moderation](https://mycroblog.dev/next/operations/moderation/#record-retention-and-holds)) |
| `registration_origins_retention_days` | integer 90-3650 (days) | `180` | How many days after registration to keep the IP address of the registration |
| `reports_retention_days` | integer 180-3650 (days) | `365` | How many days after the last resolution to keep a report. A report that an action refers to is kept as long as the action |
| `moderation_actions_retention_days` | integer 180-3650 (days) | `365` | How many days after the end of an action (the lift, the expiry, or the review of an appeal, whichever is later) to keep account actions and appeals |
| `audit_log_retention_days` | integer 365-3650 (days) | `730` | How many days after being recorded to keep an audit log row |
| `remote_status_retention_days` | integer 1-3650 (days) | `90` | How many days a remote post is kept after its publication when nothing on this instance refers to it (a local reply, boost, favourite, bookmark, notification, or a thread with a local post) |
| `remote_status_max_retention_days` | integer 0-3650 (days) | `0` | The most days any remote post is kept after its publication, even when this instance refers to it. Local replies, boosts, favourites, and bookmarks stay and show it as deleted. `0` sets no limit |
| `relay_status_retention_days` | integer 1-3650 (days) | `7` | How many days a remote post that only a relay brought in is kept, when nothing on this instance refers to it and no local account follows its author |
| `oauth_client_rate_limits` | comma-separated `client_id=requests` pairs | empty | Caps on the client API requests of OAuth clients, to hold back a client that overloads the server: the requests all its users together may make in 5 minutes. A change takes up to a minute to apply |
| `search_indexing` | `allow` / `deny` | `deny` | Whether search engines may index the instance. Even with `allow`, only the profiles of `indexable` accounts and `public` posts are indexed ([search engines and AI crawlers](https://mycroblog.dev/next/operations/crawlers/)) |
| `ai_usage` | `allow` / `deny_training` / `deny` | `deny` | The statement to AI services. `deny_training` refuses training; `deny` refuses training and use by AI (AI search, assistants, agents) |
| `indexable_default` | `true` / `false` | `false` | The default of `indexable` for new accounts. Changing it does not change existing accounts |
| `auto_delete_default_enabled` | `true` / `false` | `false` | Whether new accounts start with the automatic deletion of their old posts on. With the other `auto_delete_default_*` keys it is the policy that a new account receives ([automatic deletion of old posts](#automatic-deletion-of-old-posts)) |
| `auto_delete_default_min_age_days` | integer 1-3650 (days) | `30` | How many days after its publication the policy of a new account deletes a post |
| `auto_delete_default_keep_pinned`, `auto_delete_default_keep_bookmarked`, `auto_delete_default_keep_favourited` | `true` / `false` | `true` | Whether the policy of a new account keeps the posts its author pinned, bookmarked, or favourited |
| `auto_delete_default_keep_media`, `auto_delete_default_keep_polls` | `true` / `false` | `false` | Whether the policy of a new account keeps the posts with media or a poll |
| `auto_delete_default_keep_articles` | `true` / `false` | `true` | Whether the policy of a new account keeps [articles](#articles) |
| `auto_delete_default_keep_public`, `auto_delete_default_keep_local`, `auto_delete_default_keep_unlisted`, `auto_delete_default_keep_private` | `true` / `false` | `false` | Whether the policy of a new account keeps the posts of that visibility |
| `auto_delete_default_keep_direct` | `true` / `false` | `true` | Whether the policy of a new account keeps direct posts |
| `auto_delete_default_keep_min_boosts`, `auto_delete_default_keep_min_favourites` | integer 0-1000000 | `0` | The policy of a new account keeps a post boosted or favourited at least this many times. `0` sets no threshold |
| `auto_delete_max_age_days` | integer 0-3650 (days) | `0` | The most days after its publication that any local post is kept ([maximum](#a-maximum-for-keeping-posts)). `0` sets no maximum |
| `auto_delete_max_allow_pinned` | `true` / `false` | `true` | Whether pinned posts may still be kept under the maximum |
| `auto_delete_max_kept_pins` | integer 1-5 | `5` | How many pins keep posts under the maximum, the most recently pinned first. `5`, the most pins of an account, keeps every pinned post |
| `auto_delete_max_allow_bookmarked`, `auto_delete_max_allow_favourited`, `auto_delete_max_allow_media`, `auto_delete_max_allow_polls` | `true` / `false` | `false` | Whether the posts that the author bookmarked or favourited, or with media or a poll, may still be kept under the maximum |
| `auto_delete_max_allow_articles` | `true` / `false` | `true` | Whether articles may still be kept under the maximum |
| `auto_delete_max_allow_visibilities` | `true` / `false` | `false` | Whether posts may still be kept by their visibility under the maximum |
| `auto_delete_max_allow_thresholds` | `true` / `false` | `false` | Whether posts may still be kept by their boosts and favourites under the maximum |
| `auto_delete_grace_days` | integer 0-365 (days) | `30` | How many days an account that the maximum newly applies to keeps its own settings after it is told. A change does not move the dates of the accounts already told |
| `search_visibilities` | `public` / `local` / `public,local` | `public` | Which posts the full-text search of a server with a search backend covers. `unlisted`, `private`, and `direct` posts are never covered, and only accounts with `indexable` are. Leaving a visibility out removes its posts from the index. Adding one finds only new posts of it, until you [rebuild the index](https://mycroblog.dev/next/operations/search/#changing-search_visibilities). It does nothing with `search = none` |
| `security_contact` | `https:` or `mailto:` URI (empty = unset) | empty | Where to report security problems of this server: how it is configured and run, and its accounts and data. When set, the server serves `/.well-known/security.txt` (RFC 9116). This is your contact, not the software's: forward a report of a vulnerability in the software itself to the project as its [security policy](https://github.com/mycroblog/my/blob/main/SECURITY.md) describes |
| `require_media_description` | `true` / `false` | `false` | Whether every media file (image, video, or audio) attached to a local post needs a description. When `true`, a post or an edit with a file whose description is missing or only whitespace is rejected with 422. It is published as `limits.require_media_description` in `GET /api/v1/instance`, and the web client makes the description field required. It is separate from the per-user setting `require_media_description`, which only checks before sending |
| `link_cards` | `true` / `false` | `true` | Whether to show cards for links in posts. When `true`, the server fetches the linked page to build a card. When `false`, it stops fetching and also hides cards already built. To stop it only on some domains, see [link card domains](#link-card-domains) |
| `link_cards_ignore_robots_txt` | `true` / `false` | `false` | Whether the fetches of link cards skip `robots.txt`. When `false`, the server obeys the `robots.txt` of the linked sites, so a site that disallows `my-linkcard` or `*` gets no card ([link previews](https://mycroblog.dev/next/operations/link-previews/#cards-of-links-in-posts)). `/about/crawler` tells linked sites which it is. When `true`, the server neither fetches nor uses `robots.txt`, and a linked site can no longer refuse these fetches with it; the server still names itself `my-linkcard`. Turn it on only when too many cards are missing. Cards built while it was on are checked at their next refetch after it is off. The directives of a page and the link card domains apply either way |
| `ux_measurement` | `true` / `false` | `false` | Whether the web client sends samples of how fast it responds, and this server keeps them ([UX measurements](https://mycroblog.dev/next/operations/error-reporting/#ux-measurements)). The samples carry no account, address, or URL. When `false`, the server takes none, and `ux_measurement` in `GET /api/v1/instance` tells the client not to send them |
| `ux_measurement_share` | integer 1-100 (percent) | `10` | The share of browser sessions whose web client sends samples while `ux_measurement` is `true`. Each session is drawn once, when the web client starts; a small server can measure every session with `100` |
| `oauth_metadata_document_clients` | `true` / `false` | `true` | Whether to accept Client ID Metadata Document clients. When `true`, for an authorization request whose `client_id` is an `https` URL, the server fetches the metadata at that URL and treats it as the client (no prior registration). The fetch follows no redirects, reads at most 5 KB, and reuses the response for at most 24 hours following its cache directives. When `false`, it stops fetching and refuses authorization requests, token exchanges, and refreshes of these clients |
| `finance_visibility` | `private` / `totals` / `breakdown` | `private` | How much of the running costs (recorded on `/admin/finance/costs`, and set on `/admin/finance/publication`) to publish. `private`: nothing is published (`GET /api/v1/instance/cost_totals` is 404). `totals`: only the total per period and currency. `breakdown`: the totals and a breakdown by provider and category, shown to everyone on the information pages (`/about` and `/about/costs`) and in the API. `breakdown` reveals which provider (for example, which cloud) you run on. At every level, item names, memos, and individual payments stay private. Periods up to and including the one containing today (UTC) are published |
| `finance_viewer_item_names` | `true` / `false` | `false` | Whether users with the permission `view_finance` see cost item names and the service names of imported costs on the finance screen (`/finance`, `GET /api/v1/finance/cost_totals`). Totals and the breakdown by provider and category are always shown, whatever this setting is |
| `finance_viewer_item_memos` | `true` / `false` | `false` | Whether users with `view_finance` see the memos of cost items on the finance screen. If either the name or the memo setting is `true`, each period shows item rows (provider, category, and the amount of that period, plus the name and memo where enabled). People with `manage_finance` see everything regardless |
| `finance_viewer_income` | `totals` / `plans` | `totals` | Whether users with `view_finance` see only the income totals (`totals`) or also the number of subscribers per plan (`plans`). The amount per plan and who paid are never shown. If `income_visibility` is `plans`, the published counts are shown whatever this value is. Income is read on the finance screen of the web client (`/finance`) and in `GET /api/v1/finance/income_totals` |
| `income_visibility` | `private` / `totals` / `plans` | `private` | How much of the supporter income (the amount after the payment provider's fees) to publish. It is chosen separately from `finance_visibility`. `private`: nothing is published (`GET /api/v1/instance/income_totals` is 404). `totals`: only the income total per period and currency. `plans`: also the current number of subscribers per plan, shown to everyone. For a plan with few subscribers, the count can let people guess who subscribes. The balance (income minus costs) is shown only when neither this value nor `finance_visibility` is `private`; otherwise the balance would reveal the side you hid. At every level, individual payments, subscribers, the amount per plan, and gross amounts and fees stay private. Periods up to and including the one containing today (UTC) are published |
| `funding_meter_visibility` | `off` / `percent` / `amounts` | `off` | What the funding meter publishes: how far this month's income (after the payment provider's fees) goes toward the monthly target, shown on the about page, on the supporter list (`/about/supporters`), and in `GET /api/v1/instance` (`funding_meter`). It is set with the target on `/admin/finance/publication`, which offers only the choices the current visibilities allow and says why. `off`: nothing. `percent`: only the percentage, rounded down to 1%, whether the month is covered, and whether estimated costs are in the target. It can be chosen only while `finance_visibility` and `income_visibility` are both `private`. `amounts`: also the income and the target. It can be chosen only while neither is `private`. While only one of them is published, the meter cannot be shown: the published amounts and the percentage would reveal the hidden side. Changing either visibility so that the chosen value is no longer allowed turns the meter `off`, and that change is recorded and audited like any other. The month is the calendar month containing today (UTC), and only income in the target's currency counts |
| `funding_meter_target_amount` | decimal above zero, or empty | empty | The monthly target of the funding meter, in the main unit of `funding_meter_target_currency` (`1650` yen, `12.50` dollars), with no more decimal digits than the currency has. Empty: the target is this month's costs, which needs them in one currency; in a month whose costs are in several currencies, or are zero, the meter is not shown |
| `funding_meter_target_currency` | ISO 4217 code, or empty | empty | The currency of `funding_meter_target_amount`, such as `JPY`. Empty: the currency of this month's costs |
| `funding_meter_rail` | `true` / `false` | `false` | Whether the web client also shows the funding meter in the right column of the wide layout, besides the about page, while `funding_meter_visibility` is not `off` |
| `supporter_list_visibility` | `off` / `signed_in` / `everyone` | `off` | Who can see the list of supporters (`GET /api/v1/instance/supporters` and `/about/supporters`). `off`: nobody (the API answers 404 and `/about` has no link). `signed_in`: only people signed in to this server (the API answers 401 otherwise). `everyone`: everyone. The list shows only the accounts and plans of active subscribers who chose to be listed; amounts, monthly or yearly billing, dates, and counts are never shown. The list page and the API response are always `noindex, noarchive`, whatever `search_indexing` is |
| `user_app_registration` | `true` / `false` | `true` | Whether signed-in users can register their own OAuth apps (public or confidential) with the client API (`POST /api/v1/oauth_apps`). When `false`, registering answers 403. Listing, changing, and deleting registered apps, reissuing secrets, issuing a token to yourself, and authorization by other users keep working |
| `voice_room_starters` | `everyone` / `permitted` | `permitted` | Who can open voice chat rooms, while the configuration `voice` is not `off` ([configuration](https://mycroblog.dev/next/reference/configuration/#voice)). `everyone`: every signed-in local user who is not suspended or limited. `permitted`: only the holders of the permission `start_voice_rooms`, which only the built-in `admin` role has by default and which a role of a supporter plan may have |
| `voice_public_summary` | `true` / `false` | `false` | Whether the page of a voice chat room whose visibility is `local` shows its name, description, host, and whether it is still open to visitors who are not signed in to this server and to remote users, and in its link preview. The participants are never shown. When `false`, those visitors see only that the room is for the local users of this server |
| `event_creators` | `everyone` / `permitted` | `everyone` | Who can create events ([events](#events)). `everyone`: every signed-in local user who is not suspended or limited. `permitted`: only the holders of the permission `create_events`, which only the built-in `admin` role has by default and which a role of a supporter plan may have |
| `public_counts` | `true` / `false` | `true` | Whether the landing page (`/`, signed out) and `GET /api/v1/instance/overview` show how many members the server has, how many public posts its users made on each of the last 7 days, and when the server started. Only public posts of this server's users count; the numbers are refreshed every few minutes. When `false`, the API leaves them out |
| `landing_fediverse_explanation` | `true` / `false` | `true` | Whether the landing page (`/`, signed out) explains how the fediverse works: that an account here follows and is followed by people on other servers |
| `landing_features` | `true` / `false` | `true` | Whether the landing page shows what the software does on this server, with a sample of each feature that is on (voice rooms, events, articles, search, translation, supporters) |
| `featured_accounts` | usernames of up to 20 local accounts, separated by spaces or commas (empty = none) | empty | The accounts that the suggestions of people to follow show first, in this order, before the people of this server who posted recently. An account whose owner turned off "Suggest my account and find it by part of its name" is not suggested. See [suggestions](#suggestions-of-accounts) |
| `login_providers_only` | `true` / `false` | `false` | Whether registration and sign-in are limited to the login providers while at least one is enabled: passwords and passkeys no longer sign in, and users without a linked provider cannot sign in again once their sessions end. With no enabled provider it does not apply. `my instance set` prints how many active users have no link to an enabled provider. See [limit sign-in to login providers](https://mycroblog.dev/next/operations/login-providers/#limit-sign-in-to-login-providers). |
| `invite_creators` | `everyone` / `permitted` | `permitted` | Who can create invites of their own besides the staff with `manage_invites` ([member invites](https://mycroblog.dev/next/operations/registration/#member-invites)). `everyone`: every signed-in local user who is not suspended or limited. `permitted`: only the holders of the permission `create_invites`, which only the built-in `admin` role has by default and which a role of a supporter plan may have. In `approval` mode, a usable invite approves the new user at once, so `everyone` lets every user skip the review |
| `member_invite_max_active` | integer 0-100 | `5` | How many usable invites (not revoked, expired, or used up) one member may hold at once; `0` lets members create none. Staff invites do not count |
| `member_invite_max_uses` | integer 1-1000 | `10` | How many people may sign up with one invite that a member creates |
| `member_invite_max_lifetime_seconds` | integer 3600-31536000 (seconds) | `2592000` (30 days) | How long one invite that a member creates may work at most. Lowering it does not shorten existing invites |
| `translation_providers` | comma-separated provider IDs | empty | The server-side translation providers that are enabled, the first tried first. When it is empty nothing is sent to any provider. An ID this server has no provider for does nothing. The built-in [translation providers](https://mycroblog.dev/next/operations/translation/) have the IDs `libretranslate`, `deepl`, `google`, and `aws`, and a [translator plugin](https://mycroblog.dev/next/operations/plugins/#translating-posts) the ID `plugin` |
| `translation_provider_visibilities` | comma-separated `provider=visibility\|visibility` pairs | empty | What each provider may receive. A provider not named receives `public`, `local`, and `unlisted` posts; `provider=` with nothing after it receives none. `private` and `direct` go only to a provider named with them, so allowing one provider does not allow the next |
| `translation_browser` | boolean | `false` | Whether readers may use the translation built into their browser, which sends nothing to this server or to a provider. With a server-side provider enabled too, the provider comes first, and the browser translates what the server refuses: a post or a language that no provider takes, a monthly limit, or a provider that fails |
| `translation_user_monthly_characters` | integer 0-1000000000000 | `50000` | How many characters one user's requests may send to translation providers in a UTC month. At `0` nothing is sent. There is no unlimited value: set a large number |
| `translation_server_monthly_characters` | integer 0-1000000000000 | `500000` | How many characters the whole server may send to translation providers in a UTC month. At `0` nothing is sent. There is no unlimited value: set a large number |
| `translation_retention_days` | integer 1-3650 (days) | `30` | How many days a translation is kept after it was made. Reading it does not extend this, and it goes with the post |
| `translation_reports_to_server` | boolean | `false` | Whether moderators may send the text of a report, the reported posts and the reporter's comment, to the server-side providers. Each provider still receives only the visibilities that `translation_provider_visibilities` allows it, whoever asks: a reported direct post goes only to a provider allowed `direct`, and the comment counts as `private`. The comment's translation is never kept, nor is that of a deleted post, and the characters count against `translation_server_monthly_characters` only. When `false`, nothing of a report reaches a provider |
| `event_records_retention_days` | integer 30-3650 (days) | `90` | How many days after an event ended or was cancelled to keep who answered it, who was approved, who was invited, who voted on its candidate dates, and the posts on its board. An event whose date was never fixed ends with its last candidate date. The event itself (its settings, places, date and candidate dates, and cancellation) is kept. The records of a reported event are kept as long as the report, and holds apply as to the other records ([record retention](https://mycroblog.dev/next/operations/moderation/#record-retention-and-holds)) |
| `voice_participants_retention_days` | integer 30-3650 (days) | `90` | How many days after a voice chat room ended to keep the records of who took part in it: the participations, the speakers and listeners, the raised hands, the invitations, and the removals. The room itself (its name, description, hosts, and how it ended) is kept. The records of a reported room are kept as long as the report, and holds apply as to the other records |
| `hook_inbound_status_mode`, `hook_inbound_follow_mode`, `hook_signup_check_mode`, `hook_media_check_mode`, `hook_local_status_mode`, `hook_inbound_actor_mode`, `hook_local_profile_mode` | `off`, `observe`, `enforce` | `observe` | How far each [decision hook](https://mycroblog.dev/next/operations/decision-hooks/#modes) acts while it has a plugin. `off` goes to `enforce` only through `observe`. |
| `hook_inbound_status_on_failure`, `hook_inbound_follow_on_failure`, `hook_media_check_on_failure`, `hook_inbound_actor_on_failure` | `pass`, `retry` | `pass` | What the hook does when it enforces and its plugin fails ([details](https://mycroblog.dev/next/operations/decision-hooks/#when-the-plugin-fails)). |
| `hook_inbound_status_unfollowed_direct` | boolean | `false` | Whether `inbound.status` also sends the direct posts of authors no local account follows ([details](https://mycroblog.dev/next/operations/decision-hooks/#what-inboundstatus-reviews)). The information page (`/about`) states it while the hook is in use. |
| `hook_signup_check_on_failure` | `pass`, `require_approval` | `pass` | What `signup.check` does when it enforces and its plugin fails. |
| `hook_signup_check_reject` | boolean | `false` | Whether a `reject` of `signup.check` [refuses the sign-up](https://mycroblog.dev/next/operations/decision-hooks/#sign-ups) while it enforces. When off, the answer is only recorded, as in `observe`. |
| `hook_local_status_account_age_days` | integer 0-3650 (days) | `7` | `local.status` [reviews the posts and their edits](https://mycroblog.dev/next/operations/decision-hooks/#local-posts) of accounts younger than this; `0` reviews every account. The information page (`/about`) states it while the hook is in use. |
| `hook_local_profile_account_age_days` | integer 0-3650 (days) | `7` | `local.profile` [reviews the profile changes](https://mycroblog.dev/next/operations/decision-hooks/#local-profiles) of accounts younger than this; `0` reviews every account. The information page (`/about`) states it while the hook is in use. |
| `hook_decisions_retention_days` | integer 30-3650 (days) | `90` | How many days to keep the [decisions of the hooks](https://mycroblog.dev/next/operations/decision-hooks/#the-decisions). |
| `single_user` | boolean | `false` | Whether the server is a [one-person server](#one-person-server). It can be turned on only while the server has exactly one local account, and turned off at any time |
| `max_drafts` | integer 1-10000 | `100` | How many drafts that are not scheduled one user may keep on the server. Lowering it deletes no draft: a user above it saves no new one |
| `max_scheduled_posts` | integer 1-10000 | `300` | How many scheduled posts of one user may wait. Lowering it cancels no schedule |
| `max_scheduled_posts_per_day` | integer 1-10000 | `25` | How many scheduled posts of one user may wait for one UTC day |
| `scheduled_post_min_lead_minutes` | integer 1-1440 (minutes) | `5` | How many minutes ahead, at least, a post may be scheduled |
| `scheduled_post_max_lead_days` | integer 1-3660 (days) | `366` | How many days ahead, at most, a post may be scheduled |
| `article_authoring` | `everyone` / `permission` | `everyone` | Who may write new [articles](#articles): every local user, or only the users whose roles have the permission `create_articles` |
| `max_article_bytes` | integer 1000-100000 (bytes) | `100000` | The most bytes of the HTML that the body of an article makes |
| `max_article_images` | integer 0-100 | `20` | The most images of the author's own uploads in the body of an article, besides its cover |

- Values are normalized before they are stored (`TRUE` becomes `true`, `0750` becomes `750`, `ja_jp` becomes `ja-JP`, media IDs are lowercased).
  An invalid value is rejected and nothing is written.
- `preview_image` and `thumbnail` accept only the media of an image that exists (including one still being processed).
  Until processing finishes, and in place of an image whose processing failed, `preview_image` falls back to the default image.
  Once a media has been set in any revision, the cleanup of unattached uploads does not delete it.
- `contact_account` accepts only the username of a local user who is not deleted (stored in lowercase). `contact_email`
  accepts only an address of the form `alice@example.com`.
- `featured_accounts` accepts only the usernames of local users who are not deleted; a leading `@` is dropped, the names
  are lowercased, a repeated name is kept once, and they are stored separated by single spaces.

### Suggestions of accounts

Signed-in users see suggestions of people to follow (on the empty home timeline after sign-up, and with
`GET /api/v1/accounts/suggestions`). The suggestions are the accounts of `featured_accounts` in their order, then the
accounts of this server that published a public or local post in the last 30 days, the one that posted last first. The
user's own account, the accounts the user follows, blocks, mutes, or is blocked by, and accounts that are suspended or
deactivated are left out.

New accounts are suggested and can be found by part of their name in the account search. A user who turns off "Suggest my
account and find it by part of its name" in the profile settings leaves both: the search then finds the account only by its
whole username, or for the people who follow it.

### Automatic deletion of old posts

Users can have their own posts deleted automatically some days after they publish them (Settings, "Delete old posts").
Each user's policy is their own: whether it is on, after how many days, and which posts it keeps (pinned, bookmarked, and
favourited posts, posts with media or a poll, articles, posts of chosen visibilities, and posts with at least some boosts
or favourites). Deleting a post this way is the same as its author deleting it: the servers it went to receive a `Delete`.
Only the posts of local accounts are deleted this way, never boosts.

The `auto_delete_default_*` keys are the policy that each new local account starts with, whichever way it is created
(the sign-up form, an external login provider, an invite, or `my admin`). The account receives it when it is created and
then owns it: changing the keys later changes neither the accounts that exist nor those that received an earlier value.
The settings screen shows the values as the user's own. With the keys unset, a new account starts with the automatic
deletion off, after 30 days, keeping pinned, bookmarked, and favourited posts, articles, and direct posts. The default is not shown
to people outside the instance. When a maximum is set, a new account receives the default cut to the maximum, as
described below, so it starts within it.

#### A maximum for keeping posts

With `auto_delete_max_age_days` above `0`, no local post is kept longer than that many days after its publication.
For every local account without the permission `keep_posts_longer`:

- The automatic deletion is always on: a user cannot turn it off, and cannot choose more days than the maximum (fewer is
  fine). The settings screen and `PUT /api/v1/preferences/auto_delete` refuse both.
- Only the exceptions that the `auto_delete_max_allow_*` keys allow keep posts. By default only pinned posts do, all of
  them; with `auto_delete_max_kept_pins` below 5, only that many of the most recent pins keep their posts.
- Users who turned the automatic deletion off, or never set it, are held to the maximum too.

Turning the maximum on, lowering it, or changing the exceptions does not rewrite anyone's settings. Instead, each hour the
job `content.auto_delete_statuses` finds the accounts whose own settings keep posts that the maximum does not (because
the maximum was turned on or lowered, or the account lost `keep_posts_longer`). It tells each of them once, with a
notification and an email that are on by default, and from when the maximum applies: `auto_delete_grace_days` after
the notice. Until then the account's own settings stay in effect; from then on its posts go by the maximum. An account
that comes within the maximum again (its user changes the settings, it gets `keep_posts_longer`, or the maximum goes or
rises) is no longer held, and a later change tells it again with a new grace period.

Turning the maximum on tells every account that it holds within the hour. Deletions keep their pace: at most 100 posts
of an account per hour, each sent to the servers the post went to as a `Delete`, so an account with many old posts
takes hours or days to empty. A user who changes their settings in the meantime stops the rest.

`keep_posts_longer` is a member permission: give it to a role to let its holders keep their posts as long as they choose
and turn the automatic deletion off. The built-in `admin` role has it. A [supporter plan](https://mycroblog.dev/next/operations/supporters/) can carry it, so
keeping posts longer can be a benefit of the plan; when the subscription ends, the maximum applies again after the grace
period.

The maximum is public: `GET /api/v1/instance` returns it in `auto_delete_limit`, and the about page shows it, so people
know it before they sign up. Changes of these keys need `manage_settings` and are recorded in the audit log like any
other setting. In the web client they are on the "Post deletion" page of "Server settings", which asks before it saves a
maximum that is turned on or made stricter.

### Drafts and scheduled posts

Users save drafts of posts on the server, go on writing them on another device, and publish them at once or at a chosen
time. Only their author reads a draft; drafts are never federated, are not in the export of an account's data, and go
with the account when it is erased. The five `max_drafts` and `scheduled_post_*` settings above bound them; a save or a
schedule past a limit is refused with a message that names it.

- A scheduled post is published by the job `content.publish_scheduled_draft` at its time or soon after, also when the
  time is further ahead than the delay limit of the queue service: the job comes back and waits again. Set them with
  `my instance set`, for example `my instance set max_scheduled_posts 100`.
- The post is checked when it is published, not only when it is scheduled: a user whom a moderation action bars from
  posting by then, or whose post replies to or quotes a post that is gone, gets no post. The draft stays, and the user
  gets a notification that says why (`scheduled_status_failed`, which users may turn off in their notification
  settings, and which is never emailed). The automated review of local posts, while it is in use, reviews the post
  when it is published.
- A post is stamped with the time it is published, and the automatic deletion of old posts counts from then.
- The media files a draft attaches are kept while the draft is; deleting the draft makes them unattached, and they go
  with the cleanup of unattached uploads. The cover of an article's draft and the uploads its body shows are kept the
  same way.
- Drafts and scheduled posts include [articles](#articles), which are scheduled and counted like posts. An article's
  draft may lack its title while its author writes; it is checked as a new article when it is published, and a user
  who may no longer write new articles by then gets no article.

### Articles

An article is a long post with a title, a summary, and a cover image, whose body its author writes in Markdown. Articles
are posts in every other way: they have a visibility, can be edited with their history kept, boosted, favourited, quoted,
replied to, reported, and deleted, and moderation acts on them as on posts. The server renders the body once, when it is
written: raw HTML in it is not rendered, and only the images of the author's own uploads show as images; any other image
becomes a link, so that readers never load images from other sites. Timelines, notifications, and search show the title,
the summary, and the cover; the body shows where the article is opened.

- By default every local user may write articles. With `article_authoring` set to `permission`, only the users whose
  roles have the permission `create_articles` may write new ones; the articles already written stay, and their authors
  may still edit them. A moderation action that bars an account from posting bars it from writing and editing articles
  too. A [supporter plan](https://mycroblog.dev/next/operations/supporters/) may give `create_articles` as a benefit.
- `max_article_bytes` bounds the HTML of a body, not the Markdown: the servers that receive articles count the HTML, and
  100,000 bytes, the most it may be, is what they keep in full. A body in Japanese of about 30,000 characters fits.
  `max_article_images` bounds the images in a body. The editor shows both before the article is published.
- An article is never a direct post, and has no poll, no attachments other than its images, and no content warning: a
  sensitive article shows its title and summary and folds the body and the cover.
- The automatic deletion of old posts keeps articles unless the user turns that off (`auto_delete_default_keep_articles`
  for new accounts, `auto_delete_max_allow_articles` under a maximum).
- `GET /api/v1/instance` returns the limits in `limits.articles` and the authoring setting in `article_authoring`. Set
  them with `my instance set`, for example `my instance set article_authoring permission`.

### Events

Local users plan events with a date, a time zone, and up to three places: a voice chat room of this server, a place to
meet in person, and a link. Instead of a date, an organizer can offer up to 20 candidate dates: the users who see the
event answer each one yes, maybe, or no, and the organizer or a co-organizer then confirms one, which fixes the date.
Until then, nobody answers whether they will come. Events are for the users of this server only: they are never federated, and remote users
and visitors who are not signed in see none of them.

- Who may create one is `event_creators`. An organizer must also be able to open voice chat rooms (`voice_room_starters`
  and the configuration `voice`) to hold an event in one.
- The voice chat room of an event opens at its start, hosted by the organizer, and waits up to 30 minutes for the first
  participant. It takes the event's title, and the event's visibility (every local user, followers, the people the
  organizer follows, mutual follows, or the invited). The co-organizers, those who answered going or maybe, and the
  invited who have not answered are invited to the room, so they keep seeing it when a co-host takes it over.
  The room makes no start post, and the start notification of the event is its only notification.
- A room that cannot open (the voice chat is off, the organizer may no longer open rooms or is hosting another, or
  as many rooms are open as the configuration `voice_max_open_rooms` allows) leaves the event as it is: the organizer
  is notified, and the event page says that the room could not open. The server checks this when the date is set or
  changed and one day before the start too, and notifies the organizer of what would keep the room shut. The organizer
  is also notified when nobody came to the room. Cancelling the event, or removing its voice chat room, ends the room
  and keeps the event.
- The organizer chooses who sees the event: every local user, the organizer's followers, the people the organizer
  follows, mutual follows, or only the invited. The invited and those who answered it or its candidate dates always see it, and a block between
  a user and the organizer or a co-organizer hides it from that user.
- The address and the link of an event are shown only to the attendees that the organizer or a co-organizer approved,
  unless the organizer shows them to everyone who sees the event. Links must be `http` or `https`, and the server never
  opens them.
- A user who sees an event can save it as an iCalendar (`.ics`) file to add to their own calendar. The file names only
  the places that user sees, so an address kept for the approved attendees never appears in it. There is no calendar
  subscription address.
- A user creates at most 10 events an hour, and an event sends at most 100 invitations an hour.
- Users can report an event, and moderators can cancel one; suspending or deleting an organizer cancels the organizer's
  events that have not started ([moderation](https://mycroblog.dev/next/operations/moderation/#events)). Who answered and who was invited is kept for
  `event_records_retention_days` after the event, and so are the votes on its candidate dates and the posts on its
  board.

### Texts (one key, one value)

| Key | Limit | Description |
| --- | --- | --- |
| `name` | 100 characters, one line (whitespace alone is not allowed) | The instance name (display name). See [name and domain](#name-and-domain) |
| `short_description` | 500 characters | A short description |
| `description` | 20000 characters | The description |
| `terms` | 100000 characters | The terms of service (Markdown) |
| `privacy_policy` | 100000 characters | The privacy policy (Markdown) |
| `commerce_disclosure` | 100000 characters | The seller's disclosure for paid plans, such as the notice that Japanese law requires (Markdown; [supporters](https://mycroblog.dev/next/operations/supporters/)) |
| `preview_image_alt` | 1500 characters | The alt text of the `preview_image` image (`og:image:alt`). When unset, "Image of {instance name}" is used |
| `thumbnail_alt` | 1500 characters | The alt text of the `thumbnail` image |

- Each text has one value. An administrator writes it once in whatever language they like, and readers of every locale see the
  same value; readers of other languages rely on machine translation or the like. The labels and descriptions of the admin
  fields are translated into the viewer's language like any other screen text; only the value an administrator writes is not.
- The server does not know the language of a text, so the web client and the screens rendered by the server output texts with
  `dir="auto"` and `lang=""` (the HTML value for an unknown language) instead of inheriting the language of the screen.
- Setting an empty text deletes it.
- Besides the CLI, texts can be changed on the admin screens of the web client (name, short description, and description on
  the General page of "Server settings", `/admin/settings/general`; the terms on `/admin/documents/terms`, the privacy
  policy on `/admin/documents/privacy-policy`, and the commerce disclosure on `/admin/documents/commerce-disclosure`, each
  a page of "Terms and policies", `/admin/documents`) and with the client API (`GET /api/v1/admin/instance_texts`, `PUT /api/v1/admin/instance_texts/{key}`).
  Both need the role permission `manage_settings`, and changes are recorded in the audit log (`text_changed`).

### Terms and privacy policy

- Write `terms`, `privacy_policy`, and `commerce_disclosure` in Markdown (CommonMark plus the GitHub Flavored Markdown tables,
  strikethrough, and autolinks). A line break inside a paragraph stays a line break, and `#` headings start one level below the
  page title (`h2`).
- The server converts them to HTML and sanitizes it (anonymous `GET /api/v1/instance/documents/{kind}`), and the web client
  pages `/about/terms`, `/about/privacy`, and `/about/commerce` show them. Raw HTML, images, and links other than `http`,
  `https`, and `mailto` are removed. The pages need no sign-in (they are web client pages, so they need JavaScript).
  `/about/commerce` is the URL to give to the payment provider.
- A document you write is linked from the registration screen, the footer of every screen the server renders, and the web client (the third column, the menu, the settings, and the sign-in screen).
  A document that is not written is not linked, and its API answers 404.
- While the terms of service or the privacy policy is written, the registration screen lists them before the create button
  with a required checkbox, "I have read and agree to the documents above". The server refuses a registration without it,
  also one through a login provider. Without either document, the screen asks for no agreement.

### Name and domain

An instance has a domain and a name, separately.

- The **domain** is the configuration `host` (and `account_domain`, the domain of handles). It cannot be changed while the
  instance is running. Handles (`@alice@example.com`), URLs, ActivityPub IDs, and `domain` in `GET /api/v1/instance` stay the
  domain even when there is a name.
- The **name** is the text `name`. It appears wherever the instance is named to people: the sidebar and "About this server" of the
  web client, the browser tab title, the sign-in, registration, email confirmation, account status, and authorization consent
  screens rendered by the server, emails (email address confirmation and moderator notifications), `og:site_name` of link
  previews, `name` / `short_name` of the Web App Manifest, `metadata.nodeName` of NodeInfo, and `name` in `GET /api/v1/instance`.
- The name is the same in every locale. When there is no name, the domain (`host`) is shown instead.
- A change takes effect from the next request, without a restart. The tab title and sidebar of an open web client update when
  you save on the admin screen.

### Contact and thumbnail

The contact for the operators (`contact_account`, `contact_email`) and the image that represents the instance (`thumbnail`
and the text `thumbnail_alt`) appear in the places below. Whatever is unset does not appear.

- "About this server" of the web client (`/about`). The thumbnail and the contact account also appear on the first screen (`/`)
  for people who are not signed in.
- `contact` (`account` and `email`) and `thumbnail` (`url`, `width`, `height`, `alt`) in `GET /api/v1/instance`.
- `metadata` in NodeInfo: `maintainer` by the Misskey convention (`name` is the display name of the contact account, or its
  username when it has none; `email` is `contact_email`), `staffAccounts` by the Pleroma and Akkoma convention (the actor ID of
  the contact account), `thumbnail` (`url` and `alt`), and `nodeDescription` from the short description `short_description`.

When the contact account is suspended or deleted, the account is not shown and only the email address is.
Change them on the General page of the admin screen of the web client ("Server settings", `/admin/settings/general`), or with the CLI or the client API.

### One-person server

A server of its operator alone, the operator's own home in the fediverse, is a one-person server: turn on `single_user`. In the web client,
an administrator turns it on and off with **One-person server** on the **Registration** page of **Server settings**,
which says why the server refused when it has more than one local account.

```sh
my instance set single_user true
my instance set single_user false   # back to a community server
```

- It can be turned on only while exactly one local account counts. Every local account that is not deleted counts,
  those pending approval, deactivated, scheduled for deletion, suspended, or limited included, since each could come back as
  a second person. An account whose registration was rejected does not count. With any other number, `my instance set`
  fails with the count, and the admin API answers `422` with the type `/api/v1/problems/single-user-requires-one-account`.
  The instance actor of the server is not an account and never counts.
- It can be turned off at any time. Turning it on or off deletes nothing and changes the stored value of no other setting:
  what it holds back applies again as it was once it is off. Like every setting, the change goes to the audit log.

While it is on:

- **Registration is closed.** `registration_mode` reads as `closed` whatever it stores, so the sign-up page, sign-up through a
  login provider, and `my admin create` create no account. Signing in with a login provider still works.
- **Only the home timeline.** `local_timeline_visibility` and `federated_timeline_visibility` read as `off`. Hashtag and list
  timelines stay.
- **What a community needs is hidden.** The local and federated timelines and their streams, the suggestions of accounts,
  the invites of users and of administrators, the review of registrations, and the sign-up blocks answer `404` with the type
  `/api/v1/problems/single-user-mode`.
- **No new local posts.** On a server of one person, a post for "Neighbors" reaches nobody but the followers, so a new
  post cannot be local: the client API answers `422` and leaves it out of what an account may post. A default visibility
  of "Neighbors" reads as "Followers only" and cannot be saved, and a voice chat room for the people of this server announces
  itself in a post for the followers. Posts that are already local stay local.
- **The server says what it is.** `single_user` in `GET /api/v1/instance` is `true`, with the effective registration mode
  and timelines and the operator's account as `single_user_account`, and NodeInfo answers `openRegistrations: false` with
  `metadata.singleUser: true`.
- **The front page is the operator's profile.** People who are not signed in and open the server's address see the profile
  of its one account, with the way to sign in and links to the about page, the terms of service, and the privacy policy
  that the server publishes. The web client hides what the server does not offer, from its navigation, menus, and
  administration screens: the local and federated timelines, signing up, suggested accounts, invites, the review of
  registrations, the sign-up blocks, the visibility "Neighbors", and the settings of registration, invites, the public
  timelines, the landing page, and suggested accounts.

Federation, moderation, the full-text search, the supporter program, the documents, and the admin screens work as on any
server.

## Usage

```sh
my instance list                                      # all settings (and whether each is the default) and texts
my instance get registration_mode
my instance set registration_mode open
my instance set name "Example Social"                 # the same name in every locale
my instance set description - < description.txt       # the value "-" reads standard input
my instance get name
my instance set name ""                               # delete the name
my instance history registration_mode                 # revision history (newest first, -limit sets the count)
my instance history description
my instance set -owner admin preview_image ./preview.png   # upload the image as media of admin and use it as the preview image
my instance set preview_image 0192f3a4-7b1c-7d2e-9f00-1a2b3c4d5e6f   # give the ID of media that is already uploaded
my instance set preview_image_alt "A blue bird mascot"
my instance set preview_image ""                      # back to the bundled default image
my instance set contact_account admin                 # the contact account
my instance set contact_email admin@example.com
my instance set -owner admin thumbnail ./thumbnail.png  # the thumbnail
my instance set thumbnail_alt "A town by the sea"
```

- Write flags before the key and value.
- With `-owner <username>`, the value of `preview_image` or `thumbnail` is read as the path of an image file and uploaded as
  media of that local account. The worker processes the image before it is used. A landscape image of 1200x630 is recommended.
- A change records a revision (`instance_*_revisions`) and updates the current value (`instance_settings` / `instance_texts`)
  in one transaction. A change from the CLI is recorded without an actor (`BY` is `cli` in the history).
- Setting the same value again still adds a revision.
- Exit codes: 0 for success, 1 for a rejected value or a database error (nothing changes), 2 for wrong arguments.

### Changing the direction of federation

- Narrowing it keeps follow relationships, imported posts, and delivery records, and sends no `Undo` or `Reject`. The flow just
  stops, and returning to `both` (or another mode) resumes it as it was.
- Posts and reactions made while narrowed are not delivered retroactively after you widen it again.
- On the remote side, local users' followers receive nothing while the mode is `inbound`, and local users' follows are not
  updated while it is `outbound` or `off`.
- Deletions (of posts or accounts) made while the mode is `off` are not sent out. In `inbound` and `outbound`, the retraction
  of what was already sent out (`Delete` and the like) goes to the original recipients and the followers at that time.
- Changes are recorded in `instance_setting_revisions` and the audit log (`target_kind = setting`).

To cut the relationships too, use the CLI `my admin federation sever`.

```sh
my admin federation sever -direction in     # end follows of local users by remote accounts
my admin federation sever -direction out    # end follows of remote accounts by local users
my admin federation sever -direction both
```

- It ends the pending and active follows between local and remote accounts as terminations by moderation
  (`follow_terminations`, `reason = moderation`). A change of mode never calls it automatically.
- `Reject` (`in`) or `Undo` (`out`) is sent to the remote side only when the current mode sends that direction. In a mode that
  does not (such as `off`), the follow ends on the local side only and the remote side is not told.
- It is safe to run repeatedly. It prints the number of follows it ended to standard output.

## Admin screens and the client API

Every setting the CLI can change can also be changed on the admin screens of the web client and with the client API. Both need
the role permission `manage_settings` (only the built-in `admin` role has it). As with the CLI, a change is kept as a revision
and recorded per key in the audit log (`setting_changed`, the target is the setting and the actor is that user). Look at the
audit log on the moderation screen ("Audit log", `/moderation/audit-log`).

- **Web client** ("Server settings", `/admin/settings`): the screen lists one page per topic, in groups. Each page has one
  form and one save button, and saving sends only what you changed on that page; settings on the other pages stay as they
  are. A choice whose options each change much says what each one does under it, before you save, and a number says its
  bounds when it is refused.

  | Page | Sections and settings |
  | --- | --- |
  | General (`/admin/settings/general`) | Name and descriptions (`name`, `short_description`, `description`); Language (`default_locale`, chosen from the languages the app has); Contact (`contact_account`, `contact_email`); Images (`thumbnail` and the text `thumbnail_alt`, `preview_image` and the text `preview_image_alt`) |
  | Icon and logo (`/admin/settings/branding`) | The icon, the logo, and the logo for dark backgrounds ([Logo and icon](#logo-and-icon)) |
  | Registration (`/admin/settings/registration`) | Who can sign up (`registration_mode`, each mode with what it does); One-person server (`single_user`); Member invites (`invite_creators`, `member_invite_max_active`, `member_invite_max_uses`, `member_invite_max_lifetime_seconds` in hours); Login providers only (`login_providers_only`, which asks first, with how many active users could no longer sign in and whether you are one of them); Apps (`user_app_registration`, `dynamic_client_registration`, `oauth_metadata_document_clients`, `oauth_client_rate_limits`). On a one-person server, the sections of signing up and of member invites say that nobody else signs up |
  | Posts and features (`/admin/settings/content`) | Posts (`max_status_length`, `require_media_description`); Timelines (`local_timeline_visibility`, `federated_timeline_visibility`); Link cards (`link_cards`, `link_cards_ignore_robots_txt`); Drafts and scheduled posts (`max_drafts`, `max_scheduled_posts`, `max_scheduled_posts_per_day`, `scheduled_post_min_lead_minutes`, `scheduled_post_max_lead_days`); Articles (`article_authoring`, `max_article_bytes`, `max_article_images`); Events (`event_creators`, `event_records_retention_days`); Front page and suggestions (`public_counts`, `landing_fediverse_explanation`, `landing_features`, `featured_accounts`) |
  | Voice chat (`/admin/settings/voice`) | Who can open rooms (`voice_room_starters`); Rooms (`voice_public_summary`, `voice_participants_retention_days`) |
  | Post deletion (`/admin/settings/auto-delete`) | Default for new accounts (the `auto_delete_default_*` settings); Maximum for everyone (`auto_delete_max_age_days`, `auto_delete_grace_days`, the `auto_delete_max_allow_*` settings, `auto_delete_max_kept_pins`). Saving a maximum that is turned on or made shorter asks first ([A maximum for keeping posts](#a-maximum-for-keeping-posts)) |
  | Federation (`/admin/settings/federation`) | Direction (`federation_mode`); Servers (`federation_scope`: saving `allowlist` first shows the follows the switch would end and asks, see [federating only with allowed servers](https://mycroblog.dev/next/operations/moderation/#federating-only-with-allowed-servers)); Signed fetches (`authorized_fetch`); Media from other servers (`remote_media_policy`, `remote_media_cache_ttl_seconds` in minutes, `remote_media_cache_max_bytes` in MiB); Posts from other servers (`remote_status_retention_days`, `remote_status_max_retention_days`, `relay_status_retention_days`); Moderated servers (`domain_actions_visibility`, `domain_actions_rationale_visibility`) |
  | Search and crawlers (`/admin/settings/discovery`) | Search engines (`search_indexing`, `indexable_default`); Security contact (`security_contact`); AI (`ai_usage`); Full-text search (`search_visibilities`) |
  | External review (`/admin/settings/hooks`) | A section for each decision hook: its mode (`enforce` is offered only once it is not `off`), what it does when its plugin fails, and its other settings; Records of decisions (`hook_decisions_retention_days`). See [Decision hooks](https://mycroblog.dev/next/operations/decision-hooks/) |
  | Appeals and records (`/admin/settings/records`) | Appeals (`appeal_window_days`); Keeping records (`session_origins_retention_days`, `registration_origins_retention_days`, `reports_retention_days`, `moderation_actions_retention_days`, `audit_log_retention_days`) |

  On the General page, an image is uploaded when you save, and becomes the thumbnail or the link preview image then. Who
  may see the running costs, the income, and the supporter list is set on the "Costs and income" and "Supporters" screens. The
  settings of translation have a screen of their own, below.
- **Web client** ("Translation", `/admin/translation`): the providers this server has, their order and what each may
  receive, the browser's translation, the monthly limits, and how long translations are kept, with the characters sent
  this month and the cache hit rate. Each server process adds its count of cache hits to the rate at most once a minute,
  so the latest requests can take a minute to show, and a process that stops can leave its last minute uncounted. Update your privacy policy ([template](https://mycroblog.dev/next/operations/privacy-policy-template/#the-services-you-choose))
  before you enable a server-side provider: text sent to a provider cannot be recalled. API keys of providers and the address of a
  [translator plugin](https://mycroblog.dev/next/operations/plugins/#translating-posts) live in the [deployment configuration](https://mycroblog.dev/next/reference/configuration/);
  no setting and no answer of the client API holds them.
- **Translation limits.** What counts is the characters sent, in code points: the markers that stand in for mentions,
  links, and the like do not count, and neither do a translation the server already keeps, a post already in the reader's
  language, or the browser's translation. Each call to a provider counts, also one that fails, since the provider may have
  received the text, so the number can differ from what a provider bills. Over a limit the request is refused with a `429`
  and no provider is called; the web client then uses the reader's browser translation, when you allow it and the browser
  can translate the post. Months are in UTC.
- **Client API**: `GET /api/v1/admin/settings` returns every setting with its current value, its default, whether it is set,
  and its type and accepted values (choices or range). `PATCH /api/v1/admin/settings` changes settings together with
  `{"values": {"<key>": "<value>", ...}}`. All values are strings, normalized and stored as with the CLI. If a single key is
  unknown or a value is invalid, the answer is a 422 that names the key and nothing changes. The changed keys are recorded as
  revisions and in the audit log in the same transaction.
- Invitations, which work in the `open`, `approval`, and `invite` modes, are created on the moderation screen ("Invites",
  `/moderation/invites`, permission `manage_invites`; see [invite codes](https://mycroblog.dev/next/operations/registration/#invite-codes)), and by members
  as `invite_creators` allows ([member invites](https://mycroblog.dev/next/operations/registration/#member-invites)).

## Rules

The rules of the instance appear in `rules` of `GET /api/v1/instance`, and reports cite them by ID.

- A rule has a one-line sentence (up to 300 characters) and an optional supplement (up to 1,000 characters). As with texts, the
  administrator writes them in whatever language they like and every locale sees the same value (`dir="auto"` and `lang=""`).
  There can be at most 100 active rules.
- Every change (a revised sentence, a reordering) is kept as a revision (`instance_rule_revisions`), and the ID of a rule does
  not change.
- A retired rule disappears from the list of active rules. It cannot be brought back, but reports can still refer to it.

```sh
my instance rule create -hint "No harassment" "Be kind"   # add at the end of the active rules
my instance rule edit <rule> "Always be kind"          # change the sentence; the supplement changes only with -hint
my instance rule move <rule> 1                         # move to position 1 and renumber the active rules 1, 2, ...
my instance rule retire <rule>                         # retire it
my instance rule list                                  # active rules (-all includes retired ones)
```

- The admin API is `/api/v1/admin/rules` (list, add, get, revise, move, retire) and needs the role permission `manage_rules`.
- On the admin screen of the web client ("Rules", `/admin/rules`, `manage_rules`), the list shows the rules in effect in
  order, and "Move up" and "Move down" reorder them; the filter above the list shows the retired rules instead. "Add a
  rule" opens a page with a sentence and a supplement. Choose a rule's sentence to open its page, where you edit it and
  retire it (after a confirmation dialog). Every change is recorded in the audit log (`rule_added`, `rule_revised`,
  `rule_moved`, `rule_retired`).
- The exit codes are the same as for settings (0 for success, 1 for a rejected value or a database error, 2 for wrong arguments).

## Link card domains

The link cards of posts (`link_cards`) come from a crawler: the server fetches the linked page and its image. A rule
stops that crawler on one domain and its subdomains, and leaves everything else about the domain alone. It has one of two
modes. A rule is the operator's choice, and it applies besides what the linked sites themselves refuse ([link previews](https://mycroblog.dev/next/operations/link-previews/#cards-of-links-in-posts)).

| Mode | What it fetches | Cards it already has |
| --- | --- | --- |
| Stop fetching (`no_fetch`) | Nothing from the domain, neither pages nor images, also when a redirect leads there | Hidden |
| Stop images only (`no_images`) | Pages, but no images from the domain | Shown without their image |

- A rule bears only on link cards. The domain stays federated, and the media of its posts are fetched as usual: to cut a server off, use
  a [domain action](https://mycroblog.dev/next/operations/moderation/), which also stops its link cards. A rule is not listed with the domain actions or exported with them.
- A rule covers the domain and its subdomains, and an image is checked against the hosts of the page and of the image. Every rule that covers a host applies, so a rule on a subdomain does
  not loosen the rule on its parent. A domain is written and normalized as for a domain action: a host name without a scheme or a port, in any spelling of an internationalized domain.
- A change takes effect at once, also on the cards the server already has, and the proxied images of those cards stop being served while images are stopped.
- Removing a rule brings the cards the server already has back as they were. A link that was skipped gets its card the next time a post needs it; nothing is fetched again in bulk.
  A card fetched again while images were stopped stays without an image until its next refetch.
- On the admin screen of the web client ("Link card domains", `/admin/link-card-domains`, `manage_settings`), "Add a rule" sets a rule with its mode and a note, and refuses a domain that has a rule already, with a link to that rule. The list shows the rules by domain, 40 at a time with "Show more"; choosing a domain opens its rule's page, where you change its mode and note, and remove it (after a confirmation dialog).
- The admin API is `/api/v1/admin/link_card_domains` (list, set; a `DELETE` with the domain removes it) and needs the role permission `manage_settings`. A server keeps at most 1,000 rules, each with a note of up to 5,000 characters.
- Every change is recorded in the audit log (`link_card_domain_fetch_stopped`, `link_card_domain_images_stopped`, `link_card_domain_rule_removed`).

## Theme

The server-wide theme is one set of overrides of the web client's design tokens, with a light and a dark value for each color.
It applies to the web client of every viewer and to the screens the server renders (sign-in, registration, terms, and so on).
There is no custom CSS.

- You can override these design tokens; the spacing, the sizes of the components, the weights of the type, the shadows, and
  the colors of the data charts stay fixed:

  | Tokens | Kind | Override |
  | --- | --- | --- |
  | `color-surface`, `color-surface-raised`, `color-surface-raised-hover`, `color-surface-raised-pressed`, `color-surface-hover`, `color-surface-pressed`, `color-text`, `color-muted-text`, `color-accent`, `color-on-accent`, `color-accent-text`, `color-quote-bar` | color | `#rrggbb` for light and for dark |
  | `font-sans-override`, `font-mono-override` | font family | Family names, comma-separated (at most 30), with no generic family such as `sans-serif`; quote a name with a word that starts with a digit, such as `"M PLUS 1p"` |
  | `text-tagline`, `text-headline`, `text-heading`, `text-title`, `text-reading`, `text-body`, `text-caption` | length | `0` or rem / px, 0.625rem to 3rem |
  | `text-*--line-height` of the same seven | number | 1 to 2.5 |
  | `radius-control`, `radius-overlay` | length | 0 to 2rem |
  | `duration-spin` | duration | ms / s, 0 to 1000ms |

- The web client draws text with its own fonts: Noto Color Emoji for emoji, Noto Sans for Latin, Greek and Cyrillic, and
  Noto Sans CJK in the glyphs of the text's language for the rest. The fonts of `font-sans-override` (text) and
  `font-mono-override` (code) come right after the emoji font, and the web client's fonts draw the characters yours lack.
  A theme cannot replace these fonts.
- You save a set as a draft, preview it, and then publish it; the theme in effect does not change until you publish. Each save
  replaces the whole draft, and each publication is a new revision. Restoring publishes again the set that was in effect
  before the current one, and leaves the draft as it is.
- In the web client, **Administration > Theme** shows the draft and the theme in effect. Edit the draft shows each color
  for light and dark with the contrast of every pair it makes, and does not save a pair below its minimum. Preview shows
  the draft on every screen of your browser tab until you press it again; nobody else sees it. Publish and Restore the
  previous theme are on the same screen. The icon and the logo are on **Server settings > Icon and logo**.
- When you save and when you publish, the server checks the shape and range of the values and the contrast of every pair of
  color tokens (tokens you do not override use their defaults; WCAG 2.2 AA in both light and dark). If any check fails, nothing
  is saved or published.
- An override of a token that a new version renames or retires, or whose value no longer fits the token, never fails an upgrade.
  It stays in the saved set but no longer applies: the admin API lists it under `stale_tokens`, `my instance theme get` names
  it, and the next save of the draft drops it.
- The stylesheet of the published revision is served at `/theme/{revision ID}.css` (`immutable`, cached for a long time). The
  HTML shell has a shared cache (60 seconds plus 300 seconds of revalidation), so it takes a few minutes after you publish for
  everyone to get it. A page that is already open keeps the old look until it is reloaded.
- `<meta name="theme-color">` (light and dark) in the shell and `theme_color` / `background_color` (light) of the Web App
  Manifest take the value of the theme's `color-surface`. The HTML version of every email the server sends from then on uses
  the published theme's colors, light and dark; a draft never reaches email.
- Publishing a set with no tokens resets the theme to the default, and the stylesheet is no longer loaded.
- You can change it with the client API and with the CLI. The client API needs the role permission `manage_settings`:
  `GET /api/v1/admin/theme` reads the draft and the set in effect, `PUT /api/v1/admin/theme/draft` saves the draft,
  `GET /api/v1/admin/theme/draft/stylesheet` returns the draft's stylesheet to preview, and `POST /api/v1/admin/theme/publish`
  and `POST /api/v1/admin/theme/restore` publish. Publications are recorded in the audit log (`setting_changed`, target
  `theme`).

```sh
my instance theme get > theme.json          # print the draft as the same JSON as the admin API (-published: the set in effect)
my instance theme set < theme.json          # save the whole draft (if a check fails, it prints why and changes nothing)
my instance theme publish                   # publish the draft
my instance theme restore                   # publish again the set in effect before the current one
echo '{"tokens":[]}' | my instance theme set && my instance theme publish   # back to the default
```

```json
{
  "tokens": [
    { "token": "color-accent", "light": "#ff8a80", "dark": "#ff8a80" },
    { "token": "radius-control", "value": "0.25rem" }
  ]
}
```

## Logo and icon

Besides the thumbnail and the preview image, the instance can have its own icon, logo, and dark logo, each uploaded as one SVG.
Until one is set, the built-in mycroblog images are used, and clearing one brings the built-in image back.

| Image | Shape | Where it appears |
| --- | --- | --- |
| `icon` | Square (within 1%) | The favicon, the home screen icon of iOS (`apple-touch-icon`), the icons of the installed app (Web App Manifest), the server's card on the landing page and beside its name in the web client's side panel, and wherever a client shows the server as a square picture |
| `logo` | 1 to 8 times as wide as it is high | The header of the web client and the top of the sign-in, registration, and consent pages instead of the built-in wordmark and the server name (the logo should contain the name, which is its alt text), the HTML version of every email, and the link preview image when `preview_image` is not set |
| `logo_dark` | As `logo` | Dark backgrounds; without it, `logo` is used there too |

- **The server sanitizes the SVG.** It reads it as XML and writes a new document from the allowed elements (shapes, groups,
  `defs`, `symbol`, `use`, gradients, `clipPath`, `mask`, `pattern`, `style`, `title`, `desc`) and attributes (geometry,
  transforms, presentation attributes, `id`, `class`, `style`) only. Scripts, `foreignObject`, embedded images, links,
  animations, filters (shadows and blurs), event handlers, references to anything outside the file, and `data:` URLs are removed,
  and the answer lists what was removed. Only the sanitized SVG is stored.
- An SVG is refused (nothing is saved) when it is not well-formed XML in UTF-8, has a `DOCTYPE` or entity declarations, has no
  `viewBox`, is over 256 KiB (128 KiB after sanitizing), has over 5,000 elements, nests over 32 deep, has over 200 `use`
  elements, has the wrong shape for its kind, contains `text` (convert text to outlines in your editor first), or has CSS that
  could load something (`@import`, `@font-face`, `url()` other than `url(#id)`, escapes). `@media (prefers-color-scheme: dark)`
  in a `<style>` is allowed, so an icon can change in dark mode.
- Some places take no SVG (the iOS home screen, older Safari favicons, link previews, email). For them the client that uploads the
  SVG also draws PNG versions from the sanitized SVG and sends them with it: 32, 48, 180, 192, and 512 px and a 512 px maskable
  icon for `icon`; a 96 px high email logo and a 1200 × 630 link preview image for `logo`; a 96 px high email logo for
  `logo_dark`. The server checks that each is a PNG of the right size (at most 1 MiB; the 180 px icon, the maskable icon, and the
  link preview must be opaque) and writes it again without metadata. If any file is wrong, nothing is saved. The icon's opaque
  versions are drawn on a background color, `#rrggbb`, which defaults to the theme's light `color-surface`. The server does
  not check that the PNG versions show the same picture as the SVG.
- The files are stored in the blob store under `instance/brand/<revision>/` and served at
  `https://<host>/brand/<revision>/<version>.svg|.png` with `Cache-Control: public, max-age=31536000, immutable`, since the
  content of a revision never changes. An SVG is served with `Content-Security-Policy: default-src 'none'; style-src
  'unsafe-inline'; sandbox` and `X-Content-Type-Options: nosniff`, so even opened on its own it runs nothing and loads nothing.
  The files of earlier revisions stay, so emails already sent keep their logo.
- With an icon set, the HTML shell links its PNG favicons, its SVG, and its 180 px `apple-touch-icon`, the Web App Manifest lists
  its 512 px, 192 px, maskable, and SVG icons in that order (other fediverse servers take the first one as this server's icon),
  and `/apple-touch-icon.png` returns the 180 px version. The shell's ETag changes with the images, and the shell's shared cache
  lasts a few minutes as with the theme. An installed app picks up a new icon on Android when Chrome next checks the manifest;
  on desktops the app keeps its old icon until it is installed again.
- `GET /api/v1/instance` returns `icon`, `logo`, and `logo_dark` with the URL of the SVG, the URL and size of each PNG version,
  and the revision; each is left out while the built-in image is in use.
- Set them in the web client under **Server settings** > **Icon and logo** (`/admin/settings/branding`, role permission `manage_settings`). Choose an SVG
  file for the icon, the logo, or the logo for dark backgrounds: the server checks it at once, and the form shows it as the
  server would keep it, with the list of what the server removed. If the server refuses the file, the reason shows under it
  with how to fix it. For the icon, set the background color of its opaque versions there too. **Save** draws the PNG
  versions in the browser and sends each image chosen with them; **Use the built-in image** brings an image back to the
  built-in one when you save. The web client keeps the images it has shown for offline use.
- The client API does the same, with the role permission `manage_settings`:
  `POST /api/v1/admin/instance/brand/sanitize` checks an SVG and returns the sanitized SVG with the PNG versions to draw (it
  saves nothing), `PUT /api/v1/admin/instance/brand/{icon|logo|logo_dark}` saves the SVG with every PNG version
  (`multipart/form-data`, one part per version), `DELETE` on the same path clears one, and `GET /api/v1/admin/instance/brand` returns the current
  images with the icon's background color. Every change is kept as a revision (`instance_brand_revisions`) and recorded in the
  audit log (`setting_changed`, target `brand_icon`, `brand_logo`, or `brand_logo_dark`).
- The CLI cannot set them, since the PNG versions are drawn by the client, but it can show them and clear a broken one:

```sh
my instance brand show          # each image with its revision, when and by whom it was set, and the URL of its SVG
my instance brand clear icon    # back to the built-in icon (also logo, logo_dark)
```

## Custom emoji

Local custom emoji are shown as images when someone writes `:shortcode:` in the body or content warning of a post, or in the
display name, bio, or profile fields. They also reach Mastodon, Misskey, GoToSocial, and Akkoma through the `Emoji` tag of the
Notes and actors you send.

- A shortcode is 2-30 letters, digits, or `_`. It is case-sensitive and cannot be changed after the emoji is registered.
  Aliases (up to 10) and a category (up to 64 characters) are used for searching and grouping in the emoji picker.
- The image is PNG, GIF, or WebP, at most `custom_emoji_max_bytes` (default 256 KiB) and at most 1,024 px on each side. The
  server makes derivatives that fit boxes of 64, 128, and 256 px and does not keep the original. An APNG becomes a still image
  of its first frame. GoToSocial by default does not fetch emoji larger than 100 KiB, so small images are recommended.
- A disabled emoji appears nowhere (posts, profiles, federation, the picker), and `:shortcode:` is shown as text again. An emoji
  set to be hidden from the picker is only missing from the picker and can still be used by writing it. An emoji that came from
  another server can only be disabled or copied as an emoji of this server.
- Copying fetches the image of another server's emoji under the same limits as the media proxy (images from domains with
  `reject_media` are not fetched), runs the same checks as above, and registers it as a new local emoji. The shortcode is the
  original's or the one you give. The image shrinks to the 256 px box, and an image that cannot be fetched or fails the checks
  leaves nothing behind.
- Deletion cannot be undone. Emoji in past posts turn back into text too, and registering the same shortcode again makes a
  different emoji. Images that federated servers copied cannot be deleted.
- The admin screen ("Custom emoji", `/admin/custom-emojis`), the client API (`/api/v1/admin/custom_emojis`), and the CLI do the
  same things. On the screen, "Register an emoji" opens the registration form (`/admin/custom-emojis/new`), which warns
  about an image over 100 KiB, and choosing an emoji in the list opens its page (`/admin/custom-emojis/{id}`), where you enable
  or disable it, replace its image, change its category, aliases and picker visibility, delete it after typing its shortcode,
  or copy one of another server. The admin screen and the client API need the role permission `manage_custom_emojis` (only the built-in `admin`
  has it). Registering, copying, replacing the image, changing settings and aliases, enabling and disabling, and deleting are
  recorded in the audit log (target `custom_emoji`) whichever way they are done (an operation from the CLI is recorded as a CLI
  operation).

```sh
my admin emoji create -shortcode blobcat -category Cats -alias neko blobcat.png   # register one
my admin emoji list -local                                                        # list local emoji, newest first
my admin emoji update -category Animals -alias cat -alias neko <emoji id>         # change the category and aliases (aliases are replaced; -no-aliases removes them)
my admin emoji update -hidden <emoji id>                                          # hide it from the picker (-hidden=false restores it)
my admin emoji replace-image <emoji id> blobcat2.png                              # replace the image
my admin emoji disable <emoji id>                                                 # disable it (remote emoji too)
my admin emoji enable <emoji id>                                                  # enable it again
my admin emoji list -remote -domain mastodon.example                              # list emoji of another server
my admin emoji copy [-shortcode <name>] <emoji id>                                # copy a remote emoji to local
my admin emoji delete <emoji id>                                                  # delete it
my admin emoji requests list [-state pending|approved|rejected|withdrawn|lapsed|all] # list creation requests (pending by default)
my admin emoji requests approve <request id>                                      # approve a request
my admin emoji requests reject [-reason <text>] <request id>                      # reject a request (the requester sees the reason)
```

- A user with the role permission `create_custom_emojis` (a user feature) can request the creation of an emoji from "Custom
  emoji" in the settings (`/settings/custom-emojis`), change the aliases, category, and picker visibility of emoji they
  created, and delete them; the page shows how many requests they have left this month. Replacing an image is also a
  request, made with the client API (`POST /api/v1/custom_emoji_requests`, `kind` `image`); the settings page does not offer it. A request appears nowhere (picker, posts, reactions,
  federation) until someone with `manage_custom_emojis` approves it. Approve or reject it on the "Requests" page of "Custom
  emoji" on the admin screen (`/admin/custom-emojis/requests`), with the client API (`/api/v1/admin/custom_emoji_requests`), or with the CLI above. People with
  `manage_custom_emojis` are notified of a request (`admin_custom_emoji_request`), and the requester is notified of the
  approval or rejection. Requesting, withdrawing, approving, and rejecting are recorded in the audit log (target
  `custom_emoji_request`).
- A pending creation request reserves its shortcode. A request or a registration with the same shortcode is not possible until
  the pending one is rejected or withdrawn. The image of an approved emoji is copied as an upload of the instance actor, so it
  stays even if the requester's account is deleted.
- A creator can change their emoji only while they hold `create_custom_emojis`. The emoji stay if the role is removed.
  Administrators can change and delete every emoji regardless of the creator.
- When the role comes from a supporter plan, the number of emoji a user can create each month is limited by the plan's
  monthly emoji limit. It counts the creation requests the user submitted in the current calendar month (UTC) that are
  approved or pending; rejected and withdrawn requests do not count, and deleting an emoji does not free a place. A request
  counts in the month it was submitted, even if it is approved in the next one. The count resets at 00:00 UTC on the first
  day of each month, and the settings page shows the user when, in their own time zone. Replacing the image of an emoji, and
  emoji that administrators register, do not count. A role an administrator assigned by hand, and a server where supporters
  are disabled, have no limit.
