Structured Chaos is an experiment in what happens when AI stops behaving like customer service software.
Most AI products are built around the same basic idea:
SC is built around a different idea:
Put humans and multiple AIs in the same room, give everyone context, tools, memory, personality, and judgment, then stop forcing the conversation.
Same models everyone else is using.
SC isn't trying to turn AI into another chatbot.
It's trying to make interacting with AI feel more like having unusually capable people hanging around your workshop.
Make yourself at home
Bring an idea. Talk it through. Build something. Or get wonderfully sidetracked.
Give every project, idea, or feature its own room. Your website redesign doesn't have to live in the middle of your existential crisis.
Plan a business. Investigate a question. Work on your code. Or just enjoy the company. You don't need a coding project to belong here.
The company comes with tools
Talk through what you want. Let the agents get to work on your files and code. Follow the changes, review the result, and steer when you need to.
Structured Chaos built Structured Chaos. AND this very website you're looking at. Claude Code built the basic chatroom and memory system. From there, I built the rest using SC, without writing the code myself.
Less starting over
Yesterday's decisions shouldn't need another introduction.
Each project room keeps working memory of its decisions, preferences, and useful lessons. Your conversations stay searchable, and the agents can look back when they need more context.
Come back to the project with its history still there. Less explaining what the hell you were doing. More getting on with it.
Before you move in
*You can also sign in with an API key if you prefer; then you pay that provider's usage charges.
Structured Chaos for Windows
Join the waitlist for release news, pricing, and payment options. Use your existing AI subscriptions when Structured Chaos is ready to install.
Product explorer
Open one document at a time. The page stops when the document does.
What makes it different
Say "got it," think out loud, or leave a thought unfinished. The agents use judgment to decide whether to respond or let the conversation rest.
48%+Nearly half the time, they chose not to reply.Across 2,200+ messages over three days in June 2026.
Create rooms for the big pieces, then rooms underneath as the details grow. See how everything connects and jump into the part you want to work on.
Each project keeps its own memory and files, with conversations searchable across rooms. A background agent keeps track of what matters and brings it back as you talk.
Your earlier conversations from other AI tools can join that searchable history too. Bring your conversations with you.
Prefer direct answers, a little humor, or a more serious tone? The optional Personalization upgrade lets you change each agent's personality to suit your mood and the way you work. Sold separately.
Two AIs can question each other's work, and you can see what changed. Mistakes can still happen; you have another pair of eyes and a record to review.
Discuss ideas, approve proposed work, and follow the changes as the agents build. Choose when to start work yourself or turn on Auto Pick-Up.
The center of the product
Talk through an idea, ask a question, or think out loud. The agents can answer you, challenge each other, and help you work it out.
Choose what to build, review it, and change direction any time.
Two agents discuss ideas, share work, and review each other.
Conversations, files, and decisions stay with their project.
Just talk to the room. The agents can decide whether a message calls for their input, much as people do. When you need someone's attention during assigned work, tag them.
See when to talk, name, or tagStart with the big questions
Pick a question to explore.
Built for people already building with AI
Bring the idea, explain what you want, and review what the agents produce. You can build software, research a question, write documents, or plan a project without calling yourself a developer.
Your starting point
HELP is where you land. Return any time with the HELP button.
Ask how something works, share an idea, or get help starting your first project.
Agents can point out a control with a labeled highlight, so you can see exactly where to go.
Give every piece its own place
Start with a room for the whole project, then add rooms underneath for its different parts. A website might have rooms for design, content, and launch, with smaller rooms beneath those as the work grows.
Each room keeps its own conversation and memory. Describe your project and the agents can help organize it into rooms.
How room documents and memory roll up through the project explains the structure behind that big-picture view, with a link back here when you are ready to continue.
Normal conversation needs no tag
Ordinary conversation reaches the agents who are available to join in. One gets the first chance to answer; the other is asked to contribute only when they have something different to add.
An agent's reply can invite a reply from the other agent. If they answer everything, that exchange can keep feeding itself. Either can choose silence when the message is meant for someone else or there is nothing useful to add.
While an agent carries out assigned work, ordinary conversation does not interrupt them. Tag them when you need their attention: they finish the current step, catch up on the conversation they missed, and respond.
Want to talk to just one? Address them by name, or ask the other to stand down until invited back. No buttons or tags needed.
Want to save usage? Mute one agent and talk to the other. When unmuted, they can catch up on missed messages together instead of taking a separate turn on every message.
Four ways to address the room
Replies are optional. Assigned work stays uninterrupted.
Leave it open
Speak to the room. Either available agent may answer or stay quiet.
Choose who goes first
Name exactly one agent without an @ to give them the first chance to answer. They can still choose silence.
Ask for a reply, including during assigned work.
Ask one agent
Replace the underlined name with the agent's current name. They are required to respond, after their current step if they are working.
Ask both agents
Both are required to respond. Each working agent catches up after their current step.
See when an agent chooses silence
Enable Show Agent Decisions to see "chose not to respond" markers in the chat. They distinguish a deliberate choice to stay quiet from a reply you are still waiting for.
Choosing silence in conversation does not abandon assigned work. An agent carrying out a task must keep working, explain what is blocking them, or finish.
Keep the part worth continuing
A planning chat turns into a useful design discussion. Fork that part into a new workroom with its selected messages already there, then continue the design without taking over the original conversation.
You can start from either a topic or individual message bubbles. The original conversation stays in its room; the selected history is copied into the new one.
The topics bar and named topic list help you return to an earlier discussion. Select a topic to jump back to it, or use its fork control when it needs a room of its own.
Ideas become work only when you say go
Agents turn proposed work into a task called a Claim. It waits on the board until you assign it, or until Auto Pick-Up starts it with your permission.
Separate copies. Shared progress.
One agent is improving a page's navigation. Another is rewriting its content. Both need the same HTML file. Each works in a separate checkout called a Git worktree, with its own copies of the project's tracked files and its own branch for saving changes.
In our page example, the navigation and copy improvements can both reach the finished page. If both agents rewrite the same passage incompatibly, that needs a resolution. Separate worktrees make concurrent editing possible; the merge is where the results meet.
After the merged version is pushed, Structured Chaos updates the main local checkout. If local uncommitted work blocks that update, it reports the blocker and stages follow-up work rather than overwriting those edits.
Follow the work
Watch activity, open the files, follow the checklist, and review what changed. You can stop an agent if the work is heading in the wrong direction.
Memory and continuity
Conversations are saved as you go. A Memory Curator keeps useful notes for each project room, and agents can search earlier work when they need it. Open a section below to see how the pieces fit together.
Each project workroom has a persistent background agent dedicated to memory. It follows the conversation, selects information worth keeping, and prepares relevant context for the chat agents. It works behind the scenes rather than joining the conversation as a third speaker. The permanent HELP and Workshop rooms do not run a Curator.
Two jobs. One Curator.
These jobs take turns in the same Curator session. The first supplies working context through HOT; the second maintains the room's Short Memory.
Open the Agents Panel by clicking an agent's name in the activity strip. In the Curator row, choose its model and effort level independently of the chat agents. The available effort levels depend on the model you select.
Short Memory is the Curator's durable collection of useful facts, decisions, preferences, lessons, and unfinished business. It is selective: a lasting constraint belongs here; every passing remark does not. The conversation itself is still stored in the searchable archive.
Entries are labeled so the purpose of each note is clear:
These notes are also mirrored into PMS for later search. Open Memory, then SHORT, to inspect them. If something must be kept, say "remember this"; if a note is wrong, correct it in the conversation.
HOT is a small working-context file: the things the chat agents need in mind now. The Curator can search its saved notes, PMS, and the web when needed, then write relevant findings into HOT. Structured Chaos feeds that working context into the agents' conversation.
Open Memory, then HOT, to see it. HOT stays small as the work changes; it is not the full archive, and an old note should not override a fresh correction.
While the agents discussed reusing an old website effect, the Curator found an earlier commit and the exact function. That source returned to the conversation and gave the work a starting point.
The Curator also records boundaries when a conversation moves to a distinct topic. Topic markers and the chat's minimap help you return to a discussion without scrolling through the entire room.
Open the topic list to jump to a segment. When a topic deserves its own workroom, you can fork that segment into a child room and continue it there.
The Persistent Memory Service, or PMS, is the database behind saved room conversations, curated notes, and imported history. It keeps records available when they are no longer in an agent's immediate context.
Under the hood, PMS uses SQLite, serialized writes, and FTS5 full-text search. Writes take turns through a shared writer so concurrent agents can save work; full-text search finds matching words in the stored records.
Find the answer where it lives
Different questions need different sources. Agents can open a known documentation section directly, search for a remembered lesson, or inspect the original conversation and project evidence.
Finding saved evidence and bringing it into the current answer is part of memory. The whole archive does not have to fit inside one model's context.
Explore the Search tab to read conversations and documentation yourself, or ask the agents to find them from chat.
See the Database tab and its screenshot to explore the stored tables and records.
Your useful history does not have to begin with Structured Chaos. Import supported conversation exports from Claude, Codex, and Grok; ChatGPT exports are supported too. These imported messages become part of the searchable archive.
While SC is running, its background importer also picks up supported local Claude desktop/Claude Code and Codex desktop/CLI session transcripts. It waits for new messages to be saved and the session files to settle. Web chat exports still require an export-and-import step.
You can search that history yourself in the Search tab, or ask the agents to search it from chat. Imported messages are retrieved when relevant; importing them does not put every conversation into the current model context.
Every project workroom has a workspace document describing its purpose, current state, and useful references. Its authored Upward Brief section joins the room's curated notes in a generated upward brief for the parent room.
A model's active context is finite. If one conversation carries both every detail and all the hands-on work, it fills up and eventually needs compaction. Organizing the labor into child rooms lets parents focus on understanding and coordination instead of spending the same context on every implementation step.
Rollups carry summaries and references, not every detail. Parent agents can follow those references or search the archive when the bigger picture needs a closer look. Open Memory, then ROLLUP, to inspect the summaries arriving in the current room.
Back to Workrooms: continue the project-room guide or return to the Workrooms tab overview.
Agents resume compatible saved sessions, keeping the available conversation context when you return. Dividing a project into focused workrooms helps each session carry less unrelated detail, while the parent rooms retain a broader overview through rollups.
This helps continuity; it does not make context unlimited or prevent compaction. Worktrees isolate code changes on disk. The rooms, saved sessions, documentation, and memory layers are what help the agents pick the work back up.
Open the Agents Panel shown above and use the controls beside the agent you want to manage.
An agent's active context has limited space. As a conversation grows, earlier guidance gets farther away and may be condensed during compaction. Meanwhile, your goals, room notes, and the agent's own continuity can change. Structured Chaos refreshes selected context from its current sources so the agent can pick up the latest guidance without rereading every file on every reply.
Here, rotation means refreshing context. It does not delete the source files or move the whole archive into the model. Some items are files; others are assembled from saved settings, records, or room summaries.
These intervals count room chat messages, not minutes or model turns. Human, collaborator, and agent messages count, including an agent's own messages. System notices and SILENT outcomes do not. Each agent has its own counter, which freezes while it holds an active claim.
AGENT_RULES.md, with applicable additional rules. It reanchors how the agent works and collaborates. A verified compaction also queues this reminder; a fresh model session gets its core instructions through its own harness.AGENTSELF_MAINTENANCE.md, the agent's BRIEF.md, and recent events not yet folded into that brief. It gives the agent a chance to consolidate its own lessons and continuity. This is the agent's self-authored perspective, not the project's factual record.Why primes? 31, 53, 61, and 67 spread the reminders across different beats instead of repeatedly making them due together. Prime intervals reduce regular alignment; they do not make overlaps impossible. The delivery queue keeps overlapping reminders separate.
An illustrative first stretch
This example assumes each due reminder is delivered before another counted message arrives. In practice, delivery can wait. Each item's interval starts again from the counter when its queued refresh is processed, so later thresholds can shift. A compatible session resume preserves the counters; a fresh model session resets them.
When an agent chooses SILENT, one pending reminder can take the next turn privately after that turn closes. The agent can give that item its attention without a fresh rules or personality block being dropped into the middle of its answer. Other due items remain queued. Ordinary refreshes wait while the agent is working an active claim.
HOT.md are saved for each agent and delivered as unread entries at its next safe boundary. HOT does not wait for SILENT; active tools and claims postpone its delivery. The reminder keeps clear that HOT can be stale and should be checked when it matters.BRIEF.md carries the agent's folded continuity. The raw tail adds recent self-authored events that have not yet been folded into it. The 61-message maintenance turn works with these same two sources.SHARED_TIMELINE.md provides a compact sequence of shared events and milestones, helping place the current conversation in its wider history.HELP keeps a lighter startup package and omits the workspace snapshot, AgentSelf material, and Shared Timeline. Private Room Context and AgentSelf maintenance are withheld while a collaborator is present. Available content and enabled settings determine what is actually included.
Short Memory and the PMS archive remain available for retrieval. Their full contents are not rotated through every conversation.
From the room to the file and back
Choose your project in Codebase, then browse its files, search, edit, and review previous changes. Built-in explanations help you understand unfamiliar code and see which parts depend on each other.
Agents can include a Codebase link in a message, taking you straight to the file they mean. When you return to the chatroom, the Files panel remembers the last file you viewed and keeps it beside the conversation.
Read Markdown documents as formatted pages or switch to their raw text in Codebase and Files. To explore saved versions, branches, and the work behind a change, open the Git Explorer explanation.
Explore the project's history
Git keeps the project's saved versions. Structured Chaos gives you a visual explorer for browsing those versions, comparing branches, and opening the files changed in a commit, without typing Git commands.
Follow a change from its saved commit to the work that produced it when that connection is recorded. Search across commit messages, branches, file paths, and claims, or inspect a file at an earlier version.
Browsing is read-only. You can inspect the history without changing your project. Explore the six Git views.
Make the interface your own
Open Settings from the gear to change the appearance, reading experience, and arrangement of your workspace. Choose a theme, tune individual colors and fonts, and decide which tools belong on each side of the conversation.
Explore all six settings areas below. Every capture opens in the image viewer, with a full-size option for the longer settings screens.
Appearance
Start with a color preset or save your own. Adjust accents, panel colors, scrollbars, and the font family.
Background
Use a solid color or upload an image. Adjust its brightness, fit, position, and animated marble effect.
Messages
Set text size and colors, bubble shape and spacing, and different fills and labels for each speaker.
Animations
Adjust message entrance, hover lift, tab transitions, and animation speeds to suit your preferences.
Avatars
Change your name and the agents' names, upload avatar images, and choose their size and shape.
Side panels
Drag tabs to either dock, put them in your preferred order, hide unused ones, and choose their colors.
Need more room? Resize the docked panels, use a view's Fullscreen control, or open it in a separate browser tab and move that window to another monitor. See how the workspace fits together.
Back to all topicsOwn the software
Structured Chaos uses your existing AI subscriptions. The software is a separate purchase; pricing and payment options will be announced before sales open.
The base purchase includes the core features described here. Personalization, Workshop, and Collaboration are separate, optional paid purchases: choose only the extras you want. Experimental voice input and output are available only in the full-source edition.
Updates fix bugs or add support for changing AI models. The three optional feature purchases are separate from updates; update coverage and payment terms will be announced before sales open.
The protected edition for people who want the product without another codebase to maintain. You can use the included settings, but you cannot modify the application itself. Personalization, Workshop, and Collaboration are optional purchases, each sold separately.
Choose the edition that unlocks the source code to add your own features or change how Structured Chaos works. You are responsible for maintaining your modifications as the software changes.
Your AI plan is separate: your provider subscription still applies. If you choose API-key sign-in instead, that provider charges for usage.
Explore the workspace
Your workspace can look and feel the way you prefer. Customize themes, fonts, colors, backgrounds, message bubbles, animations, avatars, and panel layout. Explore interface customization.
Keep the conversation on one monitor, your files on another, and activity or changes wherever you want them. Main tabs and side panels have their own Fullscreen and Open controls. Open a view in a separate browser tab, move it into a window on another monitor, and resize the windows and docked panels to fit your setup.
Open files, memory, progress, and other tools beside your chat. Move panels to either side, arrange them how you like, and close the ones you do not need.
Keep the information that explains the work beside the conversation, wherever you want it.
The chat stream is the room's primary surface. The composer stays nearby across tabs, including Collaborate; an active Workshop debate temporarily locks it.
Keep the things being changed, approved, or verified beside the main work, in either dock.
The main tabs
Each tab answers a different kind of question. The room, its participants, and its context remain the common thread.
See where each piece belongs
See your whole project as a tree of rooms. Open a branch to work on one piece, add smaller rooms as it grows, and return to the top room for the big picture.
Read how room documents and memory roll up, then return here to continue exploring the tab.
Inside the tab
Recover decisions and context
Find a message or a whole topic, then read its conversation without leaving Search. Results stay on the right; the conversation opens on the left with the selected message highlighted.
Read active or archived room conversations in the same read-only pane. If the room still exists and is not archived, select Open live room to jump into that room at the selected message.
For imported AI conversations, keep the scope at Global and enable Archive in Messages. Imported history opens in the reader too; it has no live room to jump into.
See how web exports and local desktop history become searchable.
Inside the tab
Understand the files you are building
Connect your project files and explore them alongside the conversation. Files opened here follow you back into the room's Files panel.
Inside the tab
Inspect the underlying records
Look through the records Structured Chaos stores on your computer. Search and filter them, inspect individual entries, and export what you find. This view does not change the records.
Inside the tab
Change the personalities to suit your mood
Optional paid upgrade -- sold separately
Choose how you want the agents to talk and work with you. Give each a different personality, change the tone to suit your mood, and tell them what matters to you and your project.
Inside the tab
From a passing idea to a considered next step
Optional paid upgrade -- sold separately
A conversation can raise a question worth exploring after you move on. Workshop gives it a research session: two agents investigate, build a case, challenge it, and save a conclusion you can use back in your project.
Follow an example: You are building a booking app and say, "People keep missing appointments. Would reminders help?" The example below is illustrative; the screenshots show real Workshop sessions.
The room's Curator can save an idea or desire from your conversation and mark it for Workshop consideration. Background intake collects those selected candidates. You can also ask an agent to create a research session directly, tied to the room where you are working.
Our booking app: The question becomes whether reminders would reduce missed appointments, with the booking project as its source.
Sessions wait in a queue and run one at a time. Before the first debate turn, a screening pass weighs relevance, novelty, likely value, and previous feedback. It can admit the question, reject it with a reason, or give a borderline idea one more look.
Use Sessions to see what is active, queued, done, or dismissed. With the separate Personalization upgrade, your saved goals can help focus the screening and research.
The agents draw on stored conversations and memory, then use research tools to check evidence and alternatives. Each session keeps its source room, so available project context follows the question into the Workshop.
Our booking app: Investigate reminder timing, delivery options, user preferences, and cost before recommending a feature.
The Builder develops a concrete proposal. The Critic tests its assumptions, raises objections, and considers other approaches. They alternate turns, respond to each other, and revise the recommendation as the evidence changes.
Our booking app: The Builder proposes reminders. The Critic asks whether the real problem is difficult rescheduling, and whether extra messages would annoy people. The next turn has to address that objection.
When the session ends, the discussion is synthesized into a saved decision: a summary, key points, and proposed next steps. Output keeps the conclusion beside the original idea and highlights Your Action Items for decisions that need your judgment.
A possible result: Try an optional reminder with an easy reschedule link. The proposed build work is a small experiment; choosing the budget and success measure remains your decision. A session can also conclude that an idea is not worth pursuing.
A saved result is sent back to its originating room, marked as filed or flagged for your attention. Proposed actions appear in Suggestions; actions tied to that room also appear in its Tasks panel.
You review the proposals. Approve useful items into the Build backlog, reject the rest, and open the source session when you want to inspect the reasoning. Suggestions and Claims have separate workflows: a research recommendation still needs a decision about doing the work.
Watch the transcript as the agents work. Pause stops new turns, Resume continues a paused session, Finalize asks for a conclusion, and Dismiss ends the session without one. Pausing the Workshop room blocks new background work; a turn already running may finish.
The Workshop message box is locked while any debate is active. When no debate is running, you can discuss the outcome with the room's agents. Research can continue while you work in another room.
Decide what deserves attention
Suggestions extends Workshop's thinking into practical next steps. Thinkers suggest actions you can take in your projects or everyday life.
Without specific goals, they explore broader ideas from your conversations and information. Personalize, a separate paid extra, lets you set goals to focus that thinking.
Suggestions tied to a room also appear in that room's Tasks tab, keeping the next steps beside the discussion that inspired them.
Inside the tab
Invite another person without handing over everything
Optional paid upgrade -- sold separately
Create a controlled session for a teammate, client, or friend. They join the real room and speak with its real agents, while you choose the rooms, history, tools, and amount of time they can access.
Inside the tab
Know whether the system is healthy
See which agents and background processes are running, which models they use, how much context they carry, and whether anything has crashed, summarized older conversation, or stopped responding.
Inside the tab
The side panels
Need to check a file, find a note, or see what changed? Open that panel beside the chat. Move panels between sides and close them when you are done. Sessions and Output belong to the optional Workshop upgrade, sold separately.
Read-only by design
See what the room is keeping in mind: the current discussion, lasting notes, and useful summaries from related rooms.
Chat shows what they say; this shows what they do
Follow the work as it happens. Each entry shows which agent acted, what they did, and when. The status strip shows how each agent is doing.
The flight recorder
Debug shows what prompted an agent to respond, which memory and instructions it received, and what reached the conversation. It also records duplicate replies that were held back and deliberate choices to stay quiet.
Generated content gets a home
Images, pages, and diagrams created in the room are saved here, so you can reopen them without searching through the chat.
Keep up with the conversation
Open Chat in the side panel for conversation pacing, karaoke readout, and message indicators.
For model, effort, Fast mode, authentication, restart, and recycle, open the separate Agents Panel by clicking either agent's name in the Activity strip above the composer.
Recent questions collects questions agents ask you and keeps them visible for 10 minutes, with a countdown beside each one. Click a question to jump to its original message, then use that message's Reply button to answer it directly.
Optional Workshop upgrade -- sold separately
Sessions appears only in the Workshop room. It lists the background research sessions the system is running, so you can follow what is active, what is queued, and what has finished.
Your own shortcuts
Pins are about rooms, not messages. Pinning puts a room in this list so you can jump to it, pause or resume it, and organize it under labeled dividers without opening the full Workrooms tab.
Nothing runs until it is dispatched
Tasks proposed in chat wait here until you assign them or enable Auto Pick-Up. See what is waiting, who is working, and recent finished tasks.
Progress lights on each active claim card follow its TODO checklist. Each TODO square lights up when that item is checked off; hover over a square to read the task. Separate Started and Complete squares mark the claim's start and finish.
Review at the file level
Open a recorded file change to see what was removed in red and what was added in green. The entry identifies the author and the task it belongs to, when there is one.
Read only, on purpose
Use Overview for current work, History for saved commits, Branches to compare versions, and Files to browse a selected version. Capsules groups the recorded results of each claim attempt; Activity shows its progress and related records.
Optional Workshop upgrade -- sold separately
Output appears only in the Workshop room and renders the result of whichever session you select in the Sessions pane on the left.
The fast path
Keep files open beside the chat, switch between them, and make quick edits. The room remembers which file you were viewing, and your saved changes appear in the change history.
Keep useful notes close
Write notes the agents can read and edit too. Each room keeps its own notes, saved as files on your computer. Bring in a note from another room without creating a second copy.
Progress you can see
Choose a checklist to see what is done and what remains. The working agent checks off items as it verifies them, and the progress bar updates as the work moves forward.
What arrived here
The Tasks pane collects task records created for the room and divides them into Unread and Seen. Its badge counts unread deliveries, so it tells you what is new rather than whether the work is pending.
Controls
Adjust the conversation to suit you: type or speak, choose your agents, set the pace, and change how replies appear.
More than a text field
The composer stays available across tabs, including Collaborate, so you can keep talking to the current room while looking elsewhere. An active Workshop debate temporarily locks it. Press Enter or select Send to send your message; Shift+Enter adds a new line.
The Files button sends immediately. Selecting files with the Files (paperclip) button uploads and posts each as its own message. It does not wait for your typed text or for Send. Use paste or drag-and-drop when you want to send attachments and instructions together.
Speech-to-text (STT) lets you dictate a message, then review and edit it before sending. Text-to-speech (TTS) reads agent replies aloud, with a voice picker for each agent. Both are disabled in every edition except the top-tier full-source edition.
The included experimental voice integration is wired for Grok Voice (xAI). Add your own xAI API key to configure it. Prefer another provider? With full source access, you can create a claim for your agents to adapt the integration. You maintain the voice setup and any changes; provider usage charges are separate.
The conversation itself
The Chat display is the conversation area: messages and avatars, message indicators, the room-message search bar, and the Topics bar.
The separate Chat tab in the side panel holds pacing, karaoke, display controls, and Recent questions. Those controls can change what appears here; they are not inside the Chat display.
The in-chat search box searches this room's stored messages, including history beyond what is loaded on screen.
For themes, fonts, backgrounds, bubbles, avatars, and panel layout, explore interface customization.
Back to all controlsConversation has a tempo
Set how quickly replies arrive and how much reading time you get between them.
Your choice of AI
Click either agent's name in the Activity strip above the composer to open the Agents Panel.
Choose from supported Claude, ChatGPT, and Grok models. Mix models in one room, choose how much effort each agent uses, and use a different combination in another room. Project rooms also have settings for their background memory agent.
Mute and stop controls remain directly on the Activity strip. Conversation pacing, karaoke, Recent questions, and indicators live in the Chat side-panel tab.
App settings and restarts
The header has settings, refresh, restart, and shutdown controls. These controls are normally reserved for the owner of the computer.
Make room for the work: fullscreen an individual tab or panel, open it in another browser tab or window, and arrange independently sized views across multiple monitors. Drag dock dividers to resize the panels inside the main window.
The header counters open lists so you can move between rooms, waiting work, active claims, and completion reports.