Nothing matches that. Try a different word.
Rin Chat Guide
Settings map#
Every page and setting in Settings, listed from the app itself so it never goes out of date. Click one to open Settings right at it. Rin reads this same list when you ask her where something is.
Connections
LLM Providers: Where replies are generated.
- This provider
- Type: prefills the URL
- Label: what you call it
- Base URL: usually ends in /v1
- API key: never included in exports
- Max parallel requests: for a local model on one GPU — leave empty for hosted APIs
- Response timeout: how long to wait with nothing from the model before giving up, first token included; raise it for a local model that reads long prompts slowly, 0 never gives up
Image Providers: Where pictures are generated.
- Connection
- Type: which image backend this talks to
- Name: how this provider appears in the picker
- Server URL: where the server is listening
- API key: stays on this device and is never included in exports
- Model groups: your own sections in this provider's model list — shared by every generation that uses it
- Model
- Preset: whatever you saved in your own SwarmUI
Voice Providers: Reading replies aloud, and dictating into the composer.
- Chatterbox (local voice cloning)
- Start Chatterbox with Rin Chat: starts this variant whenever Rin Chat opens
- Variant
- Precision: full precision is slightly clearer; half precision uses about 1 GB less graphics memory.
- Exaggeration: how dramatic the read is; 0.5 is neutral
- Pacing: lower is slower and more deliberate; the fix for speech that is too fast
- Temperature: how much one reading differs from the next; lower is more consistent
- Free graphics memory when idle: unloads the model after this long without speech; loading it again takes about seconds
- Voice provider
- Label: what you call this pairing, e.g. Nano voice + local ears
- Read-aloud (TTS): what turns the reply into speech
- Voice: from the voices Windows has installed
- Pause between sentences: extra silence after each sentence; 0 is the engine's own pacing
- Expressiveness: how much the delivery varies — lower is flatter, higher is more animated; blank uses the voice's own value (usually 0.667)
- Rhythm variation: how much the TIMING varies — lower is metronomic, higher is loose; blank uses the voice's own value (usually 0.8)
- Compute device: what runs the model
- Language: each language is its own download
- Quality: Higher quality is clearer and a little slower, still faster than real time
- Base URL: OpenAI, or any local server speaking the same API
- API key: optional for local servers
- Model: which voice engine the server runs
- Speed: 1 is the voice's own pace; 0.5 to 2
- Collapse stretched words: Joooohn!!
- Dictation (STT): what turns your speech into text
- Whisper model: downloads once on first use, then works offline
- Dictation language: a code like en or ja; blank detects it each time you pause
- Where voice is offered
- Read aloud: a speaker button on every message; auto-read toggles per chat from the chat header
- Read my messages too: when a chat auto-reads, your messages are read as you send them, in your persona's voice
- Dictation: a mic button — uses the built-in Windows recognizer
Remote Access: Use this chat from a phone or tablet on the same network.
- Server
- Serve to my network: only while Rin Chat is open and unlocked
- Port: the port this computer listens on
- Keep character art for 3 days: instead of one hour.
- New address found: this computer's network changed, for example Tailscale just connected.
- On your network: open this on the other device, or scan the code with its camera.
- VPN: from a private VPN — works anywhere, but only on devices joined to it
- This computer only: works only in a browser on this computer, never on a phone.
- Certificate fingerprint: check this if you want to be certain it is this computer
- PIN
- Choose a PIN: at least 4 characters — digits are easiest on a phone.
- PIN entry: the number pad, or the keyboard for a PIN with letters in it.
- Ask for it again after: counted from the last time the device was used, so it won't interrupt a conversation
- Forget the PIN: stops serving until you choose a new one, and signs every device out
- Tray
- Closing the window hides it to the tray: off, the close button quits Rin Chat as it always has
- Start Rin Chat when I sign in: so it is already serving your phone once the computer is on
- Start hidden in the tray: no window on screen — just the tray icon
Discord bot: Let a character answer in your Discord server through a bot you create.
- Bot
- Bot token: stored encrypted on this computer and never shown again
- Discord developer portal: where your bot's token, intents and default name live
- Pause: stops answering and disconnects the bot
- Servers: the servers the bot has been invited to
- Invite link: adds the bot to a server you manage, with only the permissions it needs
- Replies
- Per-user cooldown: a person's messages inside this time after their last answered one are ignored
- Generation config: which profile the bot writes with: its model, provider and how much chat history it sends.
- Reply length cap: max response tokens for bot replies, never more than the generation config allows; 0 keeps the config's
- Expression pictures: a character with expression images can show one under a reply when the moment calls for it, at most 2 per channel every 10 minutes.
- Debug log: the bot's requests, replies, errors and connection events from the last day, without its token or any keys, to send with a bug report
- Pictures
- Image provider: The site's managed image API can't draw for the bot.
- Per-user picture cooldown: a person's next picture waits this long after their last one
- Picture prompt prefix: the bot's own style, put before every picture's prompt instead of your chats' default style and common prefix; blank uses those
- Channels
- Server identity: the bot has one name and picture in this server; only a channel with its own identity looks different
- Appearance in: posts with the character's own name and picture
- Pictures in: Pictures need an image provider.
- Spoiler drawn pictures in: pictures the bot draws here are blurred until someone clicks them
- Spoiler expression pictures in: the character's expression pictures posted here are blurred until someone clicks them
- Image provider in: a channel can draw with its own provider, such as one per checkpoint
- Picture prompt prefix in: used instead of the bot's prefix here, for this channel's LoRAs and style; blank uses the bot's
- History in: what the character remembers of this channel: read it, or delete single messages, in the chat itself
- What characters remember about people
- Remember people: off: replies neither see nor write any notes; what is stored stays until you clear it
- Character: whose notes to show
- Memory note: what the character is told about the person it is replying to, and how to keep notes about them.
Chat
Generation Configs: A config is a complete setup: which provider and model to run, and the samplers to run it with.
- Samplers
- Temperature: randomness
- Frequency penalty: 0 = off
- Presence penalty: 0 = off
- Top K: 0 = off
- Top P: nucleus
- Repetition penalty: 1 = off
- Min P: 0 = off
- Top A: 0 = off
- Seed: -1 = random
- Typical P: 1 = off
- Tail-free (TFS): 1 = off
- Smoothing factor: 0 = off
- DRY multiplier: 0 = off
- DRY base: ≈1.75
- DRY allowed length: ≈2
- XTC threshold: 0 = off
- XTC probability: 0 = off
- Mirostat mode: 0 = off
- Mirostat tau: ≈5
- Mirostat eta: ≈0.1
- Stop sequences: one per line; the reply is cut when any of them appears
- Other settings
- CFG scale: 1 = off (local backends)
- Trim incomplete sentences: drops a dangling partial sentence at the end of a reply
- Tool / function calling: lets the model fire a card's model-triggerable generators; some local servers reject the field
- Negative prompt: used with CFG scale above 1; local backends only
- Reset this config: samplers and prompt dials back to their defaults; provider, models, context size and tools are kept
- Model
- Config name: how this setup appears in the chat header
- Provider: the endpoint that answers
- Model: pick a provider above first — the model list comes from it
- Fallback model: retried once, automatically, if the model above is overloaded
- Max context tokens: history sent; blank = no cap
- Max response tokens: the ceiling on one reply
- Reasoning
- Effort: how hard the model thinks, and whether it thinks at all.
- Show thinking as a block: fold the thinking into a collapsible block, using the markers below.
- How to send it: gateways disagree on the field name.
- Hide reasoning: ask the provider to let the model think but not send the thinking back.
- "Don't think" tag: appended to your last message when effort is None.
- Also send chat_template_kwargs: sent automatically to GLM and Qwen 3 at effort None, since they ignore reasoning_effort.
- Thinking prefix: marker the thinking starts with
- Thinking suffix: and what it ends with
System Prompts: What the model is told before your conversation starts.
- Response language
- Language: what the AI writes in, lorebook keywords included — blank keeps the prompts' own language
- Character replies too: off keeps roleplay in the card's language while everything else follows the setting
- Config
- Config name: how this bundle appears in the chat header
- Context template (advanced)
- Template: {{system}} {{wiBefore}} {{generalDescription}} {{appearance}} {{corePersonality}} {{behaviorRules}} {{speechStyle}} {{backgroundHistory}} {{worldSetting}} {{dialogueExamples}} {{wiAfter}} , {{if x}}…{{/if}}
- Start over: reset to the built-in layout, or clear the box to use it directly
- Main system prompt
- Main system prompt: what the model is sent
- Let a character override this: when a card carries its own system prompt, send that instead of the baseline above
- Compare with a card: this config's prompts beside a card's, with the lines they share marked
- Jailbreak prompt
- Jailbreak: what the model is sent
- Author's Note
- Use in chats without their own note: applies to every config; a chat's own note still sends when this is off
- Note: what gets injected
- Depth: how many turns up from the end it sits
- Role: whose turn it arrives as
- Impersonation prompt
- Impersonation prompt: what the model is sent
- Continue nudge
- Continue nudge: what the model is sent
- Regenerate nudge
- Regenerate nudge: what the model is sent
- Start reply with
- Start reply with: what the model is sent
- Include names
- Send speaker names: some models keep speakers straighter this way; most chat models don't need it
- Group chat prompt
- Group prompt: what the model is sent
- Adventure Mode GM prompt
- Core: the directives the app parses; always sent, never editable
- Ruleset: how a roll is read: targets, modifiers, damage dice, rests, hp at zero
- Ruleset layer
- GM style: the craft: when to roll, pacing, judging spoken words, using the campaign plan
- GM style layer
Personas: Who you are in a chat.
- Your persona
- Default persona: used by every new chat that has no persona of its own
- New persona: opens a blank persona in the editor, to describe who you are in a chat
- In the library
- Hide personas from All cards: persona cards still show under Personas and in their folders
Memory & Context: What the model still remembers from earlier, and how much of it fits.
- Excerpt size (characters): how much verbatim chat goes into one recalled excerpt
- Excerpts per section: fewer is more detailed and more model calls
- Summary budget (characters): a cost on every single request — turn this down first if context is tight
- Chunk size (characters): smaller matches more precisely; larger carries more context
- Query from last N messages: how much of the recent chat becomes the search query
- Score threshold (%): how close a match has to be before it comes back
- Messages to inject: each one costs tokens on every turn
- Never recall the newest: only matters when no context budget is set; doesn't change indexing
- Install from files: if the automatic download fails
- Chat memory
- Rewrite automatically: when off, the summary never updates on its own and the model keeps relying on the older version
- Model: summarizing does not need your best roleplay model — a cheaper, faster one does it well
- Summary prompt: sent when the model writes the summary; {{char}} and {{user}} resolve.
- Reset the summary prompt: back to the built-in wording
- Scene tracker
- Scene tracker: shows the scene button and the Scene tab in a chat
- Send it every: how often the note is included in a turn
- Scene tracking prompt: what the note should track — inventory, injuries, a clock, a relationship meter.
- Reset the scene prompt: back to the built-in wording
- Long chats
- Send only the newest buttons: choices, question boxes and forms in older replies stay on screen but aren't sent to the model, which saves tokens; the newest set is always sent, so the model knows what it offered
- Prompt caching
- Cache-friendly prompt order: keeps the start of the prompt the same from turn to turn so the provider can reuse it, which is faster and cheaper; triggered lore and changing notes move closer to the end of the chat.
- Recall past messages
- Remember what scrolls out of context experimental: on for every chat; any chat can opt out from its own Memory drawer
- Embeddings from: where the vectors come from
- Local model: changing it re-embeds every chat
- Provider: which endpoint is asked for embeddings
- Embedding model: charged per token, and not covered by a flat-rate chat subscription
- Build a memory tree: recommended — it is what answers questions about when something happened
- Match on words as well as meaning: finds the exact word — a password, a price, a name — that meaning alone cannot
- Where to inject: where the memory tree lands; at a depth caches better but may confuse the order of events
- Depth: how many messages back from the end
- Injection template: must contain {{text}} or nothing is injected — saving without it keeps the previous template
- Reset the template: back to the built-in wording
- Chunk boundaries: one per line, tried in order; the last should be something that always matches
- Summarize before indexing: compresses each message before embedding so rambling ones are easier to find.
Lorebook Activation: How lorebook entries are matched and how much of the context they may take — the settings every book falls back to.
- Scanning
- Scan depth: how many recent messages keywords are matched against; an entry with its own scan depth overrides this
- Min activations: keep looking further back until this many entries fire — 0 leaves the scan at the depth above
- Max depth: how far back that search may go; 0 allows the whole chat
- Recursive scan: a fired entry's text is checked for other entries' keywords, so one entry can pull in another
- Max recursion steps: how many levels deep that chain may go; 0 is no limit
- Include names: scan “Name: message” instead of the bare message, so a key can match on who is speaking
- Matching
- Case-sensitive keys: “tavern” stops matching “Tavern”
- Match whole words: “ash” no longer fires on “ashes”
- Use group scoring: inside an inclusion group, pick the entry with the most keyword hits rather than rolling for one
- Budget & order
- Context share: the most of the context window lore may fill; 0 removes the share limit
- Budget cap: a hard token ceiling on top of that; 0 is none, and the smaller of the two applies
- Insertion order: whether the character's own lorebook or the shared ones go in first
- Warn when lore is cut: say so in the chat when the budget left entries out, instead of dropping them silently
Automation: What the app does on its own: retrying a weak reply, finishing a cut-off one, messaging you first, and filling in your name and notes on new cards.
- New cards
- Your creator name: the author the Creator field starts with; leave it empty to keep filling it in yourself
- Your creator notes: what Public Creator Notes start with, such as a short bio and your links; shown to anyone you share the card with, never sent to the model
- Fill them in on new cards: off: new cards start empty, and the Insert yours button beside an empty field is how you add them
- Auto-swipe
- Auto-swipe bad replies: re-rolls instead of leaving a poor reply in place
- Minimum length: characters below which a reply counts as too short (0 = off)
- Max attempts: how many re-rolls before it gives up and keeps the last one
- Blacklist: one phrase per line — a reply containing any of them triggers a re-roll
- Auto-continue
- Continue truncated replies: picks up a reply that ran out of tokens mid-sentence
- Max continuations: how many times one reply may be carried on
- Proactive messaging
- Quiet hours: a window when no character will message you first
- From: when the quiet window starts
- Until: when it ends
Web Search: Let replies look things up on the web, Wikipedia or a character's Fandom wiki, and read links you paste.
- Web search
- Web search: look things up before a reply when it helps
- When to search: Auto lets the model decide every turn; trigger phrases search only when your message asks for it; links in your messages are read either way
- Trigger phrases: one per line; a message containing any of them is searched
- Search provider: where searches go
- Searches left today: the subscription includes searches a day
- NanoGPT API key: each search costs about $0.006 from your NanoGPT balance
- Serper API key: from serper.dev
- Brave Search API key: from the Brave Search API dashboard
- SearXNG address: your instance's address; JSON output must be enabled in its settings.yml (search: formats: json)
- Model that decides: one short call per message; on the subscription the default is the Fast model
- Read the top pages: also fetch this many result pages and add a short extract from each; slower
- Canon lookups
- Wikipedia: the language code of the Wikipedia to read, such as en or de
- Fandom: set each character's wiki from a chat's Session info menu, under Fandom wiki
- Links in your messages
- Read pasted links: fetch the pages linked in your message
- Links per message: links past this many are left alone
- Characters per page: a longer page is cut to this length
- How results reach the model
- Depth: messages above your latest one; 0 puts the results right after it
- Role: whose turn the results appear as
- Remember searches for: the same search in the same chat reuses its results this long, so a regenerate costs nothing
Images: What gets asked for when a picture is generated: its style, the prefixes added to every prompt, and when one is drawn.
- Prompt style
- Prompt style instruction: leave blank for the built-in wording; does not affect a prompt you type yourself
- Starting points: a wording to edit, rather than a mode to switch on — each of these replaces the box above
- House style
- Default style: goes first in every picture drawn in a chat, and is the style the card image generator starts on
- Common prompt prefix: added to every prompt the app writes; one you type yourself is left exactly as written
- Negative prompt: what must never appear — applied even to a prompt you wrote, since a negative says what is unacceptable
- Include your persona's look: adds your persona card's image prompt prefix after the character's, and its negative prefix to the negative, in pictures of your persona and of the whole story
- When a picture gets drawn
- Show the prompt before generating: fixing a prompt costs nothing; rerolling a wrong picture costs a generation
- Generate when a message asks for a picture: "send me a picture of yourself" draws one after the reply — it will not fire on a message that merely mentions a photograph
- Character can send pictures: the character may add a picture of its own to a reply when the scene calls for it; each one costs a generation
- Wait between pictures: a picture the character sends sooner is skipped, with a button to draw it anyway; 0 means no wait
- Let the character read generated images: includes the picture's prompt in context so the character can react — another few hundred tokens every turn afterwards
Card Forge: How Card Forge works with you while it builds a card.
- Card Forge engine
- Card Forge engine: New: when you improve a field, edit several, or fix a grade finding, a separate writer call gets only what it needs and returns just the text.
- Messages to the guide
- Smart commands: reads each message first and turns "make example 3 spicier" into an edit of exactly that field.
- How it reads your message: Built-in needs no model: it goes by numbers and field words, with the local embedding model (23 MB, downloaded once) for the rest — instant and offline, but it is unsure more often
- Model for smart commands: a small fast model is plenty; empty = the same one as the forge
- Warn when the model thinks anyway: a notice in the forge when its model spends tokens on hidden thinking it was asked to skip
- Card grading
- Model for grading: empty = the Forge's session model there, and the active config in the card editor.
Customization
Theme & Colors: Colors and the overall look.
- Background: palette color
- Sidebars & inputs: palette color
- Cards & panels: palette color
- Hover & badges: palette color
- Alt row hover: palette color
- Title bar: palette color
- Text: palette color
- Muted text: palette color
- Borders: palette color
- Highlighted borders: palette color
- Buttons & accents: palette color
- Button borders: palette color
- Links & hover text: palette color
- Your messages: palette color
- Character replies: palette color
- Selected items: palette color
- Chat tabs: palette color
- Action buttons: palette color
- Action gradient: palette color
- Success: palette color
- Danger: palette color
- Warning: palette color
Sizing & Layout: The interface language, how big everything is, and which buttons sit on the top bar.
- Language
- Interface language: follows your system's language unless you pick one
- Apply the change: the interface switches when Rin Chat reloads — unlock again afterwards unless you're remembered
- Sizing
- UI scale: zooms the whole interface — nav, buttons, labels — for big or hi-dpi screens
- Message font size: how big the text in the chat transcript is on this computer
- Preview: how a message reads at that size
- Motion: animations in the app and in document styles on this device; Match system follows its reduce motion setting
- Animate Rin: off: she stands still in her resting pose, beside the Guide, floating and in Settings
- Grow text boxes to fit: card editor and chat sidebar fields size themselves to what's in them instead of scrolling
- Saved box sizes: a height you drag is remembered per card and overrides the setting above
Chat Display: How the transcript looks and reads.
- Automatic tags
- Tag cards you edit: when you change an imported card's content, and once for edits you'd already made
- Tag cards with new art: when you replace an imported card's image
- Expand the Thinking block by default: it stays collapsed while the model is still thinking, either way
- Message layout
- Your messages: which side your own messages sit on
- Message buttons: where copy / edit / branch / delete appear on hover
- Timed choices: let a card's timed choices count down and pick for you when time runs out
- Lorebook words: underline the words that activated a lorebook entry; hover or tap one to see which
- Card chat styles: let a character card restyle its chat window; one card's style can be turned off from that chat's Session info menu
- Character avatars: portrait beside each character message
- Your avatar: portrait beside your own messages
- Avatar size: how big avatars start in a chat; click an avatar in a chat to switch that chat
- Character expressions: where a card's expression sprites show — a chat can switch it with the button on the sprites
- Choosing the expression: how the sprite is picked from each reply — a card can always set one with [[expression: joy]]
- Auto-close italics: stops a dangling asterisk bleeding across the rest of a message
- Wrap long lines in code blocks: off, a code block keeps each line whole and scrolls sideways; plain text a model puts in a block always wraps
- Sending & scrolling
- Press Enter to send: Shift+Enter inserts a newline; off, send with Ctrl/Cmd+Enter
- Continue asks for instructions: a plain click on Continue opens the instructions box; Shift-click continues straight away
- Impersonate asks for instructions: a plain click on Impersonate opens the instructions box; Shift-click drafts straight away
- When a reply finishes: where the chat scrolls to once streaming ends
- Chat titles
- AI chat titles: show the name-with-AI button on a chat
- Auto-name new chats: after this many messages, using the cheap model
- New chat name format: date and time used when a chat is not AI-named
- Card library
- Notes on card tiles: which set of notes a tile shows
- Fall back to the other kind: when the notes you chose are empty
- Hide chats from hidden folders in All chats: chats with characters in a folder hidden from All cards; All chats can still show them for one visit
- Notes shown first in the editor: which note a card opens on — the swap button still switches the one you are looking at
- Always open the card panel for editing: in a chat, skip the read-only view and its Edit button
- Always render HTML in notes: skip the per-card question — a note's HTML and images show straight away.
- Character names
- Library shows: the Display name or the Name on tiles, lists, search, pickers and the card editor's title
- Chat shows: the Display name or the Name on messages, members, chat lists, titles, tabs and notifications
- Export uses: the Display name or the Name that exported files and bundles are named after; the card inside keeps both names
Quick Actions: One-tap replies, regenerate prompts, and the AI edit and restructure presets.
Shortcuts: Everything key- or mouse-bound, in one place.
- Regenerate last reply: Re-roll the most recent reply
- Previous swipe: Step back through the reply's alternatives
- Next swipe / new one: Step forward, or generate another alternative
- Edit last message: Open the last message in the inline editor
- Continue last reply: Extend the reply where it stopped
- Jump to composer: Put the cursor in the message box
- Find in chat: Search this chat's messages for text
- Show shortcuts: Open the shortcut list
- Mouse
- Go back: the same action as the on-screen Back button
- Clipboard history
- Built-in clipboard picker: off — the app uses only the native OS clipboard, and right-click shows the system menu
- Shortcut: opens the picker in whichever text field has focus.
Macros & Regex: Text machinery: your own and find-and-replace rules applied to messages.
- Regex live tester
- Sample text: stand-in for a model reply
- Macro playground
- {{char}}: the character's name
- {{user}}: your name
- {{persona}}: the persona description
- Input: anything you would write in a prompt or a card field
- Output: what the model would actually be sent
- Global variables
- Scope of global variables: which chats read and write them
System
Account & Security: Your local account, the vault password, and how this device unlocks.
- Account
- Password: change it any time — every device asks for the new one
- Silent unlock
- Remember me on this device: unlocks silently next launch; unchecking (or signing out) clears it
- Unlock with: asks for your fingerprint, face or device PIN, and keeps you signed in here so it has something to unlock
- aicharactercards.com account
- Unlink this device: stops sync here; sign in with your account password and re-link to enable cross-device vault recovery
- Link your account: opens aicharactercards.com to sign in with Google, Discord, or your password
- Card updates, ratings and reviews
- Check my cards automatically: once a day, a while after you unlock: links cards downloaded before card IDs existed (once per card), then looks for newer versions
- Check for updates now
- Match my cards again: looks again for cards that found no match before — useful after the card was published on the site
Data & Backup: Where your data lives, and how to get a copy of it.
- Turn off everything optional: leaves only what the terms of use require
- Moving from SillyTavern
- Migrate from SillyTavern: choose the data/default-user folder inside SillyTavern
- Settings backup
- Export settings: a JSON you can keep or share
- Import settings: overwrites the matching settings on this device
- Full backup
- Export full backup: everything, for safekeeping
- Import full backup: restores onto this device; records with the same id are overwritten
- Data folder
- Location: moving it relocates everything and reloads the app
- Export folder: where exports, backups and saved pictures go.
- Encrypt images at rest: recommended — card artwork is unreadable outside the app while locked
Diagnostics: Tools for checking the app's own behavior.
- Apply a settings fix
- Undo the last fix
- Card health check
- Show health check tips: in the card editor, above the fields they refer to
- Dismissed tips: bring back the ones you have hidden
- Debug log
- Debug logging: records the last ~300 exchanges, capped and auto-expiring after 24h
- Export the log: writes a PLAIN-TEXT file that includes your prompts and chat content (keys redacted) — only share it with someone you trust
- Clear the log: discards everything recorded so far
- Get help: bug reports, questions, feature requests — the community Discord is the support channel
- Guided tours
- Tour of the app: the title bar, the library, lorebooks, and how providers and generation configs fit together
- Tour of the chat: continue, impersonate, undo, suggest, the story director and personas — it runs inside a chat
Usage: How many tokens every model call on this device has used, and what the providers say it cost.
- History
- Clear usage history: forgets every recorded request on this device; chats keep their own counts
Updates: Keeping Rin Chat current.
- This version
- Installed version: what is running right now
- Version unknown: updates go here
- Updates: This copy is updated by your package manager, so Rin Chat doesn't check for updates itself.
- Check for updates automatically: when Rin Chat starts, with a banner when there's a new version
- Check for updates: looks for a newer release and shows what changed