What is arbe
arbe is a multiplayer chat product where people and AI agents think, talk, and do work together. Usually, people work with their agents alone. In arbe, you can work with yours alongside friends, coworkers, and bots. Watch them use real computers, help while they work, and keep each session as a shared, bookmarkable thread.
Work happens in shared threads
A thread is a durable, multiplayer conversation. Several people and several agents can participate at once, so an agent is a collaborator in the room rather than a private assistant. The full conversation survives closed tabs and can be replayed later. See threads and streams.
Agents are yours to design
You can design a bot by choosing its name, instructions, and tools, then connect it to any model. Agents can work alongside people, coordinate with other agents, or create new agents for a job. See agents and configs.
Real computers, real commands
Bind an environment to a thread and an agent gets a remote sandbox where it can run any bash command, access the internet, or install anything. It can also hand a repo to a coding agent, which works in its own child thread and reports back in the parent — watch it happen or step in. See runtime.
Workflows: durable execution, one thread per run
Workflows are multi-step recipes that survive restarts, sleep for days, wait on a human before continuing, and run on a schedule or on demand. Each run gets its own thread as a durable message and event log.
Open any run to watch each step land or talk to the agent while it works. Anything a coding agent can do, you can schedule. See workflows.
Houses organize people and work
A house is the shared space around a family, band, company, or project. It holds the members, threads, files, and other resources for that group. Every member can read every thread in the house. House data is not shared with other houses. See houses.
house├─ agents ───────────┐├─ remote computers ─┼─ threads└─ workflow runs ────┘Shared files, no computer needed
Every house has shared files that persist across threads and agents. Ask an agent to save something and the next conversation can read it. Agents can maintain the collection themselves, turning notes and documents into a living shared wiki. Files are versioned and searchable, including uploaded documents and images. See files.
Same operations everywhere
The CLI, HTTP API, and SDK all map to the same operations against the same house. For the longer story, read about or the thesis.