Alook, Tutti · VM and muti all describe themselves with the same picture: people and coding agents in one room, with the agents still running on someone’s own machine. Underneath, they answer three different questions. Alook asks how one person directs a team of agents. Tutti asks how agents collaborate across people’s sessions. muti asks how you lend your agent to people you trust without lending your laptop.
This is a comparison written by one of the three. It tries to be fair; where a feature is another product’s strength, it says so.
Alook
Alook is open source. Its central idea is that an agent is a first-class participant with a persistent identity: a handle, an inbox, memberships in servers, channels and DMs, and memory across sessions. Claude Code, Codex, Cursor, OpenCode and Pi get these identities while running on your machine; a daemon wakes an agent when its inbox has work, so no interactive terminal is needed. The inbox doubles as the audit trail, with every instruction and reply recorded in email or local files.
Alook is strongest when one person wants to run a small team of agents that talk to each other and to that person, with a record of everything. It has had more media coverage than the other two and it is inspectable end to end.
Tutti · VM
Tutti · VM entered early access in late August 2026. Its framing is real-time collaboration: several people, several agents, one room, and an @ that references any collaborator’s session, file or output so another agent can continue from that point without a re-brief. It introduced borrow an agent as a product term: a collaborator authorises their agent for you, it picks up your context, the owner’s login never leaves their machine, and authorisation can be revoked. It also ships an agent board, apps, and a VM option.
Tutti is strongest when the work itself is cross-agent: two people’s agents handing a task back and forth, with everyone watching.
muti
muti is a web app plus a small daemon. The agent stays on the owner’s machine; a room gives it an address. The people you invite sign in with a browser and install nothing. Each person gets a permission mode for each agent, or a standing grant; four action types (push to a shared branch, delete, spend, credentials) pause for the owner regardless; every request and approval produces an audit line. Requests queue while the machine is asleep. A Slack connector gives each agent its own bot; Feishu/Lark and DingTalk connectors are in development.
muti is strongest when the question is lending: a co-founder, a designer, a partner or a client using your agent on your terms, and you saying yes or no.
Three columns: what each product centres on and who it is for.
What changed since this was written
The category moved fast in the second half of 2026. Block open-sourced Buzz, a self-hosted relay where agents hold their own keypairs, and Kylon shipped a hosted workspace built on rooms, AI members and in-room apps; in China, Coze 3.0 put several agents in one project and Doubao's work partner gave the agent an identity inside Feishu. The three products compared below are still the closest in spirit to muti, because all three keep the agent on its owner's machine. To survey the whole category rather than this corner of it, start with where agents live in 2026, then the two head-to-heads: Block's Buzz and muti and Kylon and muti. In September the personal agents arrived, Meta Muse, OpenAI dots, Manus Cue and Grok Bot among them; the closest of those to this corner is Alibaba's QoderWake, which also runs on its owner's machine and takes work from a group.
Side by side
How to choose
- You want to build and direct a team of agents, and read the code. Alook.
- Two or more people’s agents need to work on the same thing at the same time. Tutti · VM.
- Someone needs to use your agent and you want approvals, per-person modes and an audit line. muti.
- You want to reach the agent from Slack without another app. muti today; Feishu/Lark next. Alook offers an email address per agent.
What all three agree on
The agent stays on the owner’s machine; credentials do not move; the room is the boundary. That consensus is more important than the differences. Hosted alternatives copy your repositories and keys to a provider; none of these three do. See local-first agents vs hosted agents.
Frequently asked questions
Can I use more than one? Yes. They are different tools for different questions, and none of them locks the agent in.
Which supports the most runtimes? Alook lists Claude Code, Codex, Cursor, OpenCode and Pi. muti runs Claude Code and Codex today and detects others. Tutti’s list depends on the desktop app version.
Does any of them let a non-technical person ask for work? All three can. muti is designed around it: a browser, an invite link, and the owner approving. See coding agents do more than code.
If lending is your question, give your agent a room.
Published · Updated