Features

Everything Orbweaver does today, as a list, so you can check a claim instead of taking my word for it. It's a snapshot and will drift; it isn't updated on every release. The app itself is the reference.

Chat

The basics, plus the things I kept wanting mid-scene.

• Swipes on the latest reply. Left and right arrows with an "n / m" counter under the newest reply step through alternate versions. Stepping past the last one generates a fresh one. The arrow keys do the same when you are not typing.
• Which model wrote each swipe. Every reply remembers the model, provider and connection that produced the version you are looking at. A small icon shows it on hover, and it stays even if you later delete that connection.
• Alternate greetings. The first message gets the same arrows, so you can step through the character card's alternate greetings until you send your first turn.
• Guided reply buttons. The composer has buttons to draft your own line (in 1st, 2nd or 3rd person), generate a reply (pick which character speaks in a group), continue the last reply, or reroll. Anything you type first becomes a steer for that action.
• Corrections. Describe what to fix, flip a few toggles, and the last reply is regenerated as an out-of-character rewrite following your instruction.
• Composer menu. The sparkle menu offers plain regenerate, post without generating, recover the last steer you used, clear input, attach media, generate an image, offer choices, and undo or re-apply a continuation.
• Undo a continuation. After continuing a reply, you can undo it back to the text from before, or re-apply it, from the message menu or the composer menu.
• Message actions. Hover or tap a message to edit it in place, fork the chat from that message into a new chat, hide it from the AI, copy it, or delete it.
• Bulk delete. Turn on selection mode, tick messages, and delete them all with one confirmation.
• See the prompt a reply was made from. On any assistant reply, the person running the chat (the host) can open the prompt that version was generated with, read back from what was stored.
• Prompt preview with context budget. A Preview tab shows what the next turn will send. A bar breaks the token budget down by source (System, Cards, World info, Steering, Game state, History). Each part opens to its full text, and a diagnostics drawer sits alongside.
• Memory recall indicator. A brain icon in the chat header shows whether memory recalled anything this turn. The host can open a popover listing what was fetched, with scores and why the near-misses lost.
• Context boundary. A divider in the transcript marks where the model's memory currently cuts off. Click it to peek at the summary standing in for the older messages.
• Thinking block. Reasoning shows as a "Thinking... Ns" section that collapses to "Thought for Ns" when the answer starts. Auto-collapse is a setting.
• Chat display styles. Switch how messages look between Bubble, Flat, Document, Echo, Whisper, Hush, Ripple and Tide. The later styles use the speaker's art in different ways: bled portrait, banner, tall sticky portrait, or per-paragraph bubbles.
• Avatars and message details. Choose avatar visibility, size, shape, aspect and accent ring. Add per-message chips for timestamp, id, token count, generation time and cost, and choose whether message actions show on hover or always.
• Reply cost on demand. Click the readout on a reply to fetch its final cost from the provider (OpenRouter).
• Streaming and message settings. Enter-to-send, empty-Enter to generate or continue, auto-continue when a reply hits the length cap, auto-swipe for short or blacklisted replies, custom stopping strings, smooth streaming speed, and scroll-follow versus pin-my-message.
• Attach images and video. Drop a file on the composer, paste from the clipboard, or pick from the sparkle menu. Images and short videos upload and go with your next message.
• Next-turn line. When a reply cannot be served, a line under the composer says why and how to fix it. The model that wrote a reply can show with that message's actions.
• Slash commands and command palette. Type / in the composer for commands like /new-chat, /gallery, /imagine and /plugin, with completion. Press Ctrl/Cmd+K to jump to any chat, section or command. Use // for a literal slash.
• Chat library. Search chats by title, character or newest message, filter by character and month, and star, archive, rename or delete from each row.
• Export chats and import SillyTavern transcripts. Import one or many SillyTavern .jsonl chat files (or an Orbweaver bundle) from the Chats list. Export a chat as a full room backup, a .jsonl transcript, or plain .txt.
• Temporary chats. A Home tile starts a throwaway chat that deletes itself after a set number of hours. You choose the number.
• Home. Home opens on the room you last spoke in, with its last line and who is there, plus recent rooms and character quick-picks.
• New chat picker. A searchable picker starts a chat with one or several characters, a blank chat, or a saved roster.
• Group chat members. The Members tab lists people and characters in the room. Add or remove characters, mute one, set how talkative each is, force one to speak next, and view a character's card limited to what the room is allowed to see.
• Group behavior. The host chooses who speaks each round (Natural, Everyone in order, Round-robin, Only when I pick, or Smart, using a side model). They also pick per-speaker replies or one narrator message, per-character card scoping, and an auto mode where characters answer each other for a set number of turns.
• Shared rooms. The host creates a one-time share link or invites people by handle. Guests see a preview and join, and the roster shows who is online. The host can hand off hosting, remove people, and hide pre-join history. Members can leave.
• Emoji reactions. Members can react to messages, even to one character's part of a message. The host can let the model add reactions from present characters while it replies.
• Offer choices. The model can end a reply with numbered choices shown as buttons. Clicking one either puts it in your composer to edit or sends it, depending on your setting. Ask for one set from the composer menu, or keep it on in "This chat".
• "This chat" tab. One pane of per-chat settings. Override the main prompt, post-history or scenario for this room. Add manual injections at chosen positions and depths. Attach documents and world books. See and toggle which regex scripts run here. Answer a preset's macro questions.
• Host controls. The host can cap how many tool rounds a turn may take, set a background for this chat, and share their display regex scripts with everyone in the room.
• Sandboxed HTML cards. The model can send an HTML card that renders in a locked-down frame. Cards that run their own scripts are on by default for every character, inside that frame. An admin can switch scripts off for everyone, and each character can be set lower on its Trust tab. Untrusted output never touches the page itself.
• Character gallery. Each character has an image gallery with upload, a full-screen viewer and a "this chat" filter. Open it from Chat options or /gallery.
• Edit the app's instructions to the model. Rewrite the wording of the system instructions Orbweaver sends for summaries, memory digests, arbitration and injections. Leave one blank to use the default.
• Per-chat memory, documents and image prompt settings. Set memory recall depth and summarization, document retrieval (chunking and reranking), and the /imagine prompt templates for portrait, scene, background and caption.

Characters, lore and prompts

Cards, books and presets, with the SillyTavern import doing most of the carrying.

• Import SillyTavern character cards. Drop one or many SillyTavern PNG or JSON cards on the Characters list. They land in your library with the embedded lorebook, regex scripts, tags and alternate greetings carried over.
• Import a whole SillyTavern profile. Backup & Restore accepts a .zip, a single card, or an unzipped SillyTavern profile folder. It merges characters, chats, personas, presets, themes and world files in the background, after a check that states what the import will and will not touch.
• Export character cards. Export any character as a PNG card with the avatar embedded, or as a data-only JSON card. Use the list row menu or the open character's Actions menu.
• Characters library. Search your characters, sort them many ways, and switch between a flat view and a categorized one where tags act as collapsible folders. Filter by favorites and archived. Tag chips cycle through off, include and exclude.
• Character shelves. With no character open, the Characters pane shows shelves such as who you last talked to, who you starred, and what you added recently. Open a card or start chatting right away.
• Character editor. Edit a card with no Save button; it saves as you type. Fields cover Description, Personality, Scenario, Example messages, Creator notes, System prompt, Post-history instructions, Note at depth, Regex scripts and Provenance. The header shows total versus permanent tokens, with a greeting preview (alternates included) and a spoiler blur.
• Character side panel. Beside the editor, tabs show the character's chats, links to world books and personas, a per-card theme override, a snapshot history with one-click restore, and per-card rules for how HTML and outside media in its messages may display.
• Restore a card's embedded world book. On an imported character, pick the original card file to put its embedded lorebook back. A confirmation names what it will overwrite.
• Character and persona conversion. Turn any character into a persona from its Actions menu, or create a persona from a character in the Personas list. You can optionally swap the character and user point-of-view macros in the description.
• Personas. Each persona has a description, injection placement and depth, an attached lore book, and can be pinned as your default for new chats. The switcher marks who you're playing as, your default, and who the card calls you; a switch inside a chat changes that chat, and Everywhere carries it to new chats too. "Restamp" relabels your earlier messages in a chat to the current persona. A popover on the sidebar switches personas anywhere.
• The persona pin. {{user}} in card text (description, personality, scenario, greeting, examples and the card's lore) is the persona that opened the chat, and stays that persona through a mid-chat swap. Preset text and your own lines follow your current persona. In a shared room the host picks which player the card means. Started as my SillyTavern extension PersonaPin.
• First-run persona prompt. A new account without a usable persona must name its {{user}} before chatting. The dialog can't be closed, and it also repairs a half-created state.
• Presets with a visual prompt rack. The preset editor has Params, Prompt, Actions, Data and Transforms views. Prompt is a drag-and-drop stack of sections around a Chat History pivot, with enable switches, drill-ins, markers such as the world-info anchors, and keyboard reordering.
• Real token costs per prompt section. When the preset editor is tied to a real chat, the Prompt readout shows what each section actually costs in tokens, including conversation and world info, and drills into the exact rows. Without a chat it shows a plain placeholder instead of a fake zero.
• Prompt-cache warning. If a section above Chat History contains a macro that changes every turn, the editor warns that it breaks the history cache and names the sections to move below Chat History.
• Model-aware sampling and reasoning controls. The Params view shows only the sampling, reasoning, verbosity and stop-sequence settings the chosen model supports, plus max tokens, logit bias, and context compaction on a Claude subscription connection. You can switch which connection it describes.
• Import SillyTavern presets. Drop a SillyTavern Chat Completion preset .json and it becomes a preset, with its prompt order mapped to sections and a plain list of any fields that had no home. Orbweaver preset exports import too and merge by name.
• Preset actions, variables and macros. Beyond the prompt, a preset holds many per-action prompt templates (continue, impersonate, guided and others). It also has choice-style {{variables}} you answer when generating, your own macros with a browser of every macro the preset sees, and a list of attached regex scripts.
• Editing the built-in preset. Editing the built-in default makes your own copy and tells you so, with a choice dialog if you already have one. One preset is marked active for generation, with a toast when it changes.
• World Info books. A library of keyword-triggered lore books. Each entry has keywords, content, priority, a fire mode (auto, always, keyword), an always-include option that bypasses the budget, and optional injection at a chat-history depth with a role. Turn a book on globally, per character, per persona or per chat; the editor shows where it fires. Scan depth and token budget are settings.
• Regex scripts. A library of find-and-replace scripts with placement, depth range, macro substitution, trim-out and run-on-edit. Attach them globally or from a preset, character or room. A "Try it" tester runs the real engine, and a debugger shows what will run, in order, and why a script was skipped.
• Documents. Add files, pasted text or links (a web page, YouTube captions, a wiki article). They are indexed once, and relevant passages are pulled into chats as they happen. Attach globally, per character or per chat. A panel lists the rooms where each document is active. When an extractor upgrade ships, a banner offers to re-extract the files read with the older one.
• Saved rosters. Save a room's characters, their seat settings, the anchor persona and the enabled automation rules as a named roster. Start a new chat from one, or add one to the open room.
• Labels. One place to manage tags for characters, chats, world books, personas and presets: colors, folder type, hide-on-cards, merge, prune unused, and an inbox to accept or reject suggested tags.
• Open a character in the Refinery. From a character's Actions menu, jump into a Refinery session for that card. The overview shows a card quality score once analyzed.

Games, automation and images

The game panel, and the automation and pictures that ride along with it.

• Game mode. The host flips a switch in the chat's three-dots menu (or the Game tab) and the roleplay becomes a game. The model keeps tracked state (health, gold, quests, scene) while it writes. You can turn it off and back on without losing anything.
• Game HUD. On a game chat, the top of the side panel becomes a live display: a status strip with location, time and weather, pinned stat orbs and a coin disc. Tabs underneath cover Status, Inventory, Scene, Quests and Journal.
• Game state follows your swipes. Each reply version carries its own snapshot of the game state. Swiping or regenerating swaps the display to that version's state, and a reroll is generated from the state before the rejected version, not after it.
• Custom trackers. The host defines named trackers: a meter like Health or Grit, a text field, or a list. Each belongs either to every character or to the whole game. A short description tells the model what it means, and meters can be pinned to the status strip as orbs.
• Hand edits stick. When you edit a tracked value by hand (HP, an item, the act, the characters list), a small pin marks it as locked so the model can't overwrite it. A Release button hands control back.
• Character sheets and rulesets. Each game character has an expandable sheet with title, level, wallet, attribute values and live trackers. The host picks Freeform (prose only) or D20 (six attributes, a health meter and dice), and can add or rename attributes and set the value range.
• Dice rolls. With the D20 ruleset on, the composer's Message tools menu gets "Roll d20 / d6 / 2d6 / d100", which insert the result into your draft. Dice rolls the model makes show as a readable roll card in the thread.
• Quests. Quests are cards with objective checklists shown as a segmented ring. Done and failed quests move to an archive, and an "ACT II" progress rail tracks the campaign arc. The host can edit names, objectives and acts in place.
• Inventory. Each character has a pack showing item icon, name, quantity and location, with a party currency total and a grid or list toggle. The host can edit, add and drop items in place.
• Scene tab. See who is on stage with mood and relationship, plus location, time, weather, custom widgets and recent beats. NPCs the model invents are tracked without a card, and you can promote one to a full library character with a seat in the room.
• Journal and Marks. The Journal tab is the campaign chronicle: beats grouped by day, editable by the host, plus an archive of cards. The host can save a named checkpoint ("Mark") and later restore the game to it.
• Resync from story and Fill from card. "Resync from story" re-reads the recent story and rebuilds a drifted display; hand-locked fields are never overwritten, and you can optionally restamp your messages to your current persona. "Fill from card" has the model set up a character's sheet, starting gear, coin and implied quests from their card.
• Clickable choices. A "CYOA choices" switch (choose-your-own-adventure) makes every reply end with a set of clickable choices. A pick either drops into your composer to edit or sends immediately, and the Scene tab shows the live choice. Separately, "Offer choices" in the composer wand asks for one set on the next reply in any chat.
• Steer the plot. With Plot steering on, the composer wand gets a Story group with "Steer the plot": Natural progression, Random twist, Grounded twist, Escalate, De-escalate and Advance the act. Each is a one-shot nudge for the next reply.
• Immersive HTML cards. The model can answer with self-contained HTML cards (letters, notices, terminals) shown in a sandbox, optionally animated or scripted. Cards are archived in the Scene tab and in the Journal by the day they appeared. Older cards shrink to a stub in the prompt to save tokens.
• Hidden information. The host can teach the model to let characters lie (a standing-secrets channel) and to note events the party can't perceive (Omniscience). The host sees a "Veiled" ledger of claim, truth and origin. Members never see hidden content.
• Game console. The host's Game tab holds the table's settings: which preset the game's turns speak in (the GM voice), game macros, a steering note members never see, how many recent beats go into the prompt, and how state is recorded (inside the reply in one call, or a second cheap pass, recommended for local models).
• What this turn did. Each game reply has a collapsed footer listing the state changes the model made that turn, tied to the version you are viewing.
• Automation rules. In the host's "This chat" tab, an "Add a rule" picker offers many ready-made rules: welcome-back recap, auto-add lore, pacing nudge, illustrate the scene, call a vote, clock fires, the needle, prose audit, spotlight balance, auto-set scene background and more. There is also a rule editor for writing your own: pick the event, write a CEL condition, and list the actions in order. Each rule has a free Test dry-run, Run now, an enable switch, a log explaining why it did or didn't fire, a rate cap and its own model connections.
• Approve before the AI acts. Rules that would write lore, post a message or fix prose raise a confirmation card above the composer. The host approves or dismisses, and nothing lands until confirmed.
• Quick-reply chips. Rules like Opener chips, Offer chips after a beat and Call a vote show a row of editable suggestions above the composer. Any member can click one to send or compose it.
• Story meters. The Needle rule has a side analyst score scene tension and shows it as a dial beside the transcript, shifting the backdrop when it runs high. The Clock rule shows a segmented countdown that fills one step per beat and fires when full.
• Library-wide rules and rate limit. Settings -> Automation holds rules that watch your whole library with no chat open, such as Living library, which repaints a character portrait when the card changes. A runs-per-hour ceiling covers all of them.
• Room Activity tab. The host's Activity tab lists everything automation did in the room while you were away: auto-fires, Run-now runs, notices, plugin tool runs and confirmed suggestions.
• /imagine. /imagine opens a modal with modes (free, you, face, scene, background). The extraction modes can "Read the chat first" to build an editable prompt, showing the price before the image is generated and posted. The composer wand also offers "Imagine — modes & preview..." and an instant generate-from-typed-text action.
• Pictures in replies. A Reply pictures switch in a preset's Params lets a chat model that can draw, such as Gemini's image models, answer with pictures in the chat beside its text. A text-only model ignores it and says so. → Images
• Image viewer. Clicking a generated image in chat opens a viewer showing how it was made and what it cost. "Edit" does an image-to-image change from your instruction and saves a new image you own. "Set as background" uses it for the room.
• Automatic scene illustration and backgrounds. Rules can generate a picture of the current scene on a schedule or on a lore reveal, swap the room background to one of your own that fits, or repaint a character portrait when the card changes.
• Image prompts. A settings section edits the prompt templates and negative prompt used for portrait, scene, background and caption image generation.

Library, memory and insights

Where the library stops being a folder of PNGs you forgot about.

• Library search. One search box covers your whole library. Switch the target to find characters by meaning, past chat scenes by describing the moment, saved memory summaries, images by what they show or their caption, or plain keyword search on card text.
• Library overview. The Corpus home shows your library at a glance: a map of visual families clustered from character portraits, story themes pulled from your chats, a keyword explorer, theme drift over time, characters you invested in then dropped, characters you never played, and a list of which analysis passes have run.
• One-click understanding pass. One button reads your cards into genre, tone and pitch, builds memory for your existing chats, and groups chat summaries into story themes. A live progress bar lets you leave and come back.
• Character dossier. Opening a character in Corpus shows a card-quality read, how well the portrait matches the card, tags and keywords, similar characters, and look-alike art from your library.
• Ask a character card. Ask a question about a character ("How do they treat allies?") and get an answer drawn from your recent scenes with them, labelled Grounded or Speculative.
• Browse by genre and tone. With nothing typed, Corpus Explore lists your analyzed characters with Genre, Tone and Tag filters (each showing counts) and Recent or Name sorting.
• Scenes and memories. Click a scene or memory result to see a window of the original transcript, the generated memory summary, related rooms and alternate swipes from the same moment. A button opens that exact point in the chat.
• Semantic map. Every indexed character appears as a dot on a 2D map where nearby dots are similar, colored by genre. Click a dot to open that character.
• Archetypes, Visuals, Similarity and Compare. Whole-library analysis tabs: art and writing archetypes (clusters of cards), image traits such as art style, palette, mood and outfit, lists of the closest pairs and duplicates across characters, art and chats, and a side-by-side comparison of two characters with an optional model-written deep comparison.
• Insights dashboard. A stats page for your own usage: characters, chats, words, replies, swipes, forked chats, time spent generating, active days and streaks, generation speed, token and cache figures, which characters are rising or falling between your two most recent active months, and an all-time recap of your roleplay. → Stats and Insights
• Recompute now. Rebuilds your stats from your chat history in one press, and refuses if a rebuild is already running. An import runs the same rebuild when it finishes, and the owner can run it for every account from Jobs.
• Honest accounting. Every token, cost and cache figure is labelled Recorded, Estimated or Not recorded. A missing figure is a dash, never a zero, and imported SillyTavern turns show estimated counts with a ~.
• Per-character stats. A leaderboard of your characters sortable by replies, generation time, swipes or recency. Drill into one for its activity, costs and latency. Side tabs break usage down by model (with each model's time to first token and throughput), by day and weekday-hour heatmap, and by persona (only Personas follows the drilled character; Models and Time stay library-wide).
• Memory for long chats. A setting summarizes older parts of a chat in the background and recalls those summaries later, so the model stays consistent across a long thread. It starts off, and turning it on states the cost and which model pays. A Memory backfill job builds it for chats you already had, and the turn-on box can start it for you with its call count shown.
• Context boundary and summary. A divider in the transcript marks where the model's context begins. When older messages were compacted, it says so and lets you view the summary standing in for them.
• Refinery. Pick a character and run a model critique of each card field with a 1-10 score. Rewrite only the fields you chose using that critique and your guidance, then analyze the rewrite against your original. Nothing touches the card until you accept.
• Refinery review, versions and save-as-copy. After a rewrite you keep or reject each field one at a time, then apply what you kept (the card is snapshotted first) or save it as a copy. A Versions tab lists snapshots you can compare with the live card and restore.
• Custom Refinery output formats. Describe in plain English the structure you want the critique or rewrite to come back in. A draft format is generated, and you can preview it, test it on a real card, and save it.
• Background jobs. A settings pane to start, watch, retry and cancel background jobs over your library: indexing for search, analyzing characters, computing themes, memory backfill, finding duplicates and importing. A run that calls a model shows about how many calls it will make and asks before it starts, and so does a retry. Progress is live.
• Job schedules and tuning. Run a job nightly, hourly or weekly. Tune thresholds for duplicate detection, theme granularity and keyword co-occurrence limits.
• Import and restore backups. Drop a backup zip or a single character card (PNG or JSON), or pick an unzipped Orbweaver backup or SillyTavern profile folder. A check first states that import adds, skips anything already in your library, and files a same-named but different world book, theme or preset under a numbered name. A per-file report shows what happened.
• Export your library. Download one archive of characters, chats, personas, world info, regex scripts, documents, presets, themes, settings, tags, gallery and media, with a checkbox for each kind.
• Notification inbox. A bell in the top bar collects chat invitations, removals, host handoff requests, notices from your automations, and plugin alerts, including the first-run prompt to allow newly installed plugins. Accept, decline or dismiss from there.
• Home page. Opens on the room you were last in and lists other open rooms. It offers quick starts, temp chat, saved rosters and a Documents tile, plus a grid of jumps to every other section.

Setup, sharing and plugins

The part I over-built. It's all here anyway.

• Connections. Add a connection by picking a provider, pasting a key or a server URL, and picking a model from a searchable list. The key is stored once and never shown again. Your first chat-capable connection becomes the Chat model and asks which model runs background work, with "use the same one" as a choice.
• Supported providers. Google Gemini, OpenRouter, Anthropic, Claude subscription, OpenAI, vLLM, LM Studio, Ollama, any OpenAI-compatible URL, and a built-in on-device provider for embeddings and reranking.
• Model roles. Choose which connection each job runs on: Chat, a cheap Utility model for summaries, structured extraction and image captions, Image generation, Text embedding, Image embedding and Rerank. Each slot shows what it needs and what gets skipped if the model can't do it.
• Connection editor. Per connection: a server reachability check with an "Admit host:port" fix, extra request body fields, custom headers and response path mapping with a sample-response preview, an inspector showing the redacted request and the server's answer, prompt-cache controls, and an OpenRouter credit balance.
• Saved keys. See every key your connections reuse. Replace or revoke one and every connection using it follows.
• Themes. Pick from shipped themes (Hearth, Mocha, Light) and your own in one grid. Build a theme from every color and style setting, with a custom-CSS editor, a live preview and contrast warnings. Duplicate, export to JSON and import.
• Backgrounds. Put a wallpaper behind the whole app: pick a bundled scene, upload an image or video, or add one from a URL (fetched and stored by the server). Set the fit mode, scrim darkness and blur.
• Reading and layout tuning. Fine-tune on top of your theme: chat column width, global text scale, compact or comfortable density, flat or layered depth, line height, letter spacing, justified text, frosted glass, film grain, accent-tinted borders and a reduced-motion override.
• Settings that explain themselves. Every settings group, section and option has a plain description of what it does and what it affects. A search box with @modified, @advanced and @in: filters jumps straight to a setting. The same index is in the Cmd/Ctrl+K palette.
• Install a plugin with a consent screen. Add a plugin from a zip, folder, bundle URL or Git repo. Read the exact list of permissions it asks for, then approve or deny. Later updates that widen its reach ask again.
• Example plugins. Every new account gets ready-made example plugins installed but switched off, with no permissions granted. Examples include a research familiar, oracle deck, affinity tracker, draft polish, scene chips, story clocks, pocket arcade, keepsake camera and card atlas. One click to review and enable.
• What plugins can add. A full page in the Plugin pages section, a side panel in a chat, buttons in the message tools menu, a strip under each message, dialogs, /plugin commands and Cmd/Ctrl+K entries, and rendered cards for their model tools. The app draws all of it, so a plugin can't pose as part of the app itself.
• Plugin spending and logs. A plugin that calls a model uses a connection you choose for it per task, shown as "Not set" until you do. Each plugin has its own log panel.
• Publish a plugin to every account. An admin can publish a plugin to all accounts on the deployment or withdraw it.
• Multi-user accounts. Owner, admin and user roles. Create local accounts, reset passwords, enable or disable accounts, and list and revoke any account's sessions. An optional discreet login hides handles.
• Single sign-on. Run behind OIDC (Authentik and similar) or forward-header auth. Admins can require approval for new SSO sign-ins, and link an SSO identity to an existing local account so switching modes keeps all data.
• Share with friends. The owner presses Start sharing. The app checks the setup, starts a Cloudflare quick tunnel, and shows a link and an invite-to-room picker. On a LAN or public IP it can also get a Let's Encrypt certificate for the IP address.
• Invite links. A friend opening a room invite signs up (local) or signs in (SSO), names their persona, and lands seated in the room. An expired invite explains itself.
• Session recovery. When a login session expires, a password prompt appears over the live app. Your open chat, draft and scroll position survive.
• Deployment safety and tuning. Admins can block external media URLs, control rich HTML and interactive cards, cap generated-image size, set rate limits and login-attempt caps, and allow specific private-network endpoints. They can also tune the memory pipeline, prompt cache and structured-output strictness, refresh the model catalog, backfill card embeddings, and restart the server.
• About this install. Shows the exact version and commit, with a copy button for bug reports and a manual check for whether you're current on GitHub.
• Self-hosting. A one-line installer (install.sh, or install.cmd on Windows), then start and update scripts for every day. Or by hand: docker compose up, or pnpm install and pnpm start. The first start asks for the port and who may sign in (just me, me and friends, or SSO).
• Mobile and installable web app. On a phone the app shows one pane at a time with a bottom tab bar (Home, Chats, Characters, You), and everything else lives behind the You sheet. It can be added to your home screen and handles the on-screen keyboard.
• Workspace layout. A rail of sections with list, content and detail panes that dock, collapse or overlay depending on width, a focus-mode toggle, and a Cmd/Ctrl+K palette listing settings, plugin commands and slash commands.

Now go install it and find the one I forgot to list.