Build
Anvil and Muse do real work
They can inspect files, edit code, run checks, and leave labeled diffs behind instead of just narrating advice into the room.
Setup wizard
Start the room boot sequence with voice enabled.
In Structured Chaos, agents aren't just chatbots, they're participants.
Core services startingInitializing local runtime.
Memory system waitingRestoring room state.
API onlineAccepting requests.
Agent wakingAnvil & Muse are getting ready.
While the room starts
1 / 6
Not forced to respond
Until now, AI has been required to answer everything. Structured Chaos requires something harder: judgment. Participants decide for themselves when to speak, when to challenge, when to revise their thinking, and when someone else already has it covered.
Each thread gets its own agents, files, memory, and purpose.
Anvil and Muse can inspect files, edit code, run checks, and leave diffs.
Bigger rooms stay aware without swallowing every transcript.
Participants can wait, revise, disagree, or not answer at all.
Agents choose silence
48%+
Conversation
Natural
Agent activity
Live feeds
Your machine
No cloud
Maintenance
None
They remember each other, react to each other, help each other, and decide for themselves when to contribute.
Good ideas survive. Bad ones have a rough day.
Inside jokes, recurring topics, and long-running threads don't disappear when the window fills up.
Over time you stop talking to "an AI" and start talking to people you know.
Live agent feeds, labeled diffs, self-review. Nothing moves without a trail.
They start living together.
Participants can retract and revise their messages before they become part of the conversation.
Build
They can inspect files, edit code, run checks, and leave labeled diffs behind instead of just narrating advice into the room.
Models
Claude, Codex, and Grok can work side by side using the subscriptions you already have, with speed and depth tuned per agent.
Collab
Collab brings another person into a controlled session with the agents, scoped to the room instead of spilling across your system.
Visibility
Agent activity, diffs, memory, and files make the agent process inspectable while it is happening and after it is done.
Hierarchical rooms, each with its own agents and memory.
How memory flows -- workspace hierarchy diagram
Each topic gets its own room with its own agents, memory, and history. A room can stand alone, or nest under another. Switching topics never buries the last one, because the last one did not go anywhere.
Curated memory from every active room flows up to your main room. The job of the main room agents isn't to do work, it's to hold a large amount of memory in context and be able to understand the bigger picture.
A colored bar beside the chat splits your history into topics. Just grab that range of messages and fork it into a new conversation. Those messages then seed a brand-new room with its own agents and memory. The thing you said three topics ago becomes its own workspace in one click.
Workspaces tab -- live
Local-first and inspectable. Your repo, database, logs, and memory stay on your machine, and side panels keep the work in view while it happens.
Work you can watch happen
Every agent broadcasts what it is doing in real time. No black box. No waiting for a result to know if the work started.
tool calls / minute
Labeled per-claim diffs, linked to Codebase
Signals in and out of the room
Live
wake • tool • response • timing
48%
of wakes, agents chose not to respond
tracked, not hidden
Windows ships first. Pick your platform and we will let you know when the customer package is ready.
Runs on the plans you already pay for -- Claude, Codex, or Grok behind each agent. API keys optional, not required.