Muse, dots, Grok Bot, Viktor and Claude Tag in Slack: what changes for a team

In short. Your team's Slack can now hold two kinds of AI agents. Personal assistants, like Meta's Muse, OpenAI's dots and Grok Bot, work for one person. Shared AI teammates, like Viktor and Anthropic's Claude Tag, work for a whole channel. Both kinds are capable, and both are worth using. The question for a team is who can direct them, and what holds the rules and the results once several of them are working on the same thing. This comparison uses each vendor's own documentation, checked on October 12, 2026.

Two ways agents join Slack

The split is not ours. Anthropic's Claude Tag documentation draws it plainly: team work belongs in Claude Tag, and personal work in Cowork or Claude Code (Anthropic). The other products line up the same way.

Personal assistants work for their owner.

• dots (OpenAI). An always-on agent with its own cloud computer that can hand tasks to ChatGPT Work or Codex (OpenAI). You can message your dot in Slack, and it can join with its own Slack account (OpenAI).
• Grok Bot. Named Bots that each work on a persistent cloud computer, and your Bots can message each other and pass a task along (Grok Bot docs). When a Bot prepares a Slack message for you, it shows up as a draft you send (Grok Bot docs).
• Muse (Meta). A personal agent that runs on its own Muse Secure VM and that you talk to in the Muse app or in WhatsApp (Meta). Muse for Small Business adds Slack to its list of connectors (Meta).

Shared AI teammates work for the channel.

• Viktor. "One AI employee your whole team shares, working in the Slack channels you invite him to" (Slack Marketplace).
• Claude Tag (Anthropic). One @Claude for the organization. Anyone in a channel can tag it into a problem, and anyone can steer it by replying in the thread (Anthropic).

The personal assistant makers are moving toward teams as well. Grok Team Bots are shared Bots with their own Slack app (Grok Bot docs). OpenAI runs Team Tasks that several teammates maintain (OpenAI) and has previewed specialist dots for organizations (OpenAI).

Side by side

Every cell links to the vendor's own page. We list mechanisms, not scores.

Personal assistants: the channel is shared, the agent is not

A personal assistant in a team channel is easy to see and hard to share. OpenAI's workspace guide is explicit: only a dot's owner can direct it, and other people's direct messages or mentions do not start work for that dot (OpenAI). By default a dot responds to its owner, who can tell it to engage with others (OpenAI). A personal Grok Bot works through its owner's Slack account (Grok Bot docs). Muse is not in the channel at all. It reaches Slack through its owner's connectors (Meta).

Each one also remembers its owner, not the team. A dot shares memory with its owner's ChatGPT and keeps using the plugins its owner connected, even when it has its own Slack account (OpenAI). All of a person's Grok Bots run on one computer tied to that person's account, and the documentation advises against putting a credential there that another of your Bots should not use (Grok Bot docs). When you disconnect a connector from Muse, details it already used can stay in its memory (Meta Help Center). Where personal context stops and team work starts depends on each owner's settings.

And when the owner is away, the agent's part of the work waits for them. A teammate who picks up the task cannot direct someone else's dot.

Shared AI teammates: one vendor, one agent, one harness

Shared teammates solve the direction problem. Anyone on the team can put Viktor to work (Slack Marketplace), and anyone in a channel can steer Claude by replying in its thread (Anthropic). If someone is out, the agent is still there.

Each one is still a single vendor's agent, with that vendor's harness. Claude Tag works with the service accounts an admin sets for each channel and keeps memory per channel (Anthropic), and its custom instructions apply to Claude (Anthropic). Viktor connects to your tools on its own and remembers your processes, decisions and preferences (Viktor). A rule the team agrees on has to be taught to each of them.

They also read the room differently. When someone mentions @Claude partway into a thread, Claude reads a window of the thread with other bots' replies filtered out (Anthropic), so a reply Viktor just posted there is not part of what Claude reads. None of these products documents a way to hand work from a shared teammate to the personal assistants next to it, or back.

Put two shared teammates and three personal assistants in one Slack and a team runs five harnesses: five places where memory lives, five sets of rules and five approval flows. A person stitches them together.

What an agent-native workspace does differently

Kylon does not replace an agent's own harness. A brought Claude Code still runs its own loop and tools on its owner's computer, and a hosted agent runs on Kylon's. What the workspace adds is the layer a team needs around all of them, the company harness: who can direct each agent, what context it reads, which rules apply, which accounts it may use, what waits for a person, and where results go. That layer covers what agents do through the workspace. What a brought agent does with its own local tools and accounts stays under its own settings.

• Anyone on the access list can direct an agent. Every agent has its own member list. Members can message it, mention it and assign it work, and every agent keeps at least one human admin (Kylon docs). A new agent starts private, with only its creator on the list.
• Personal and shared agents sit in the same room. Hosted agents run on Kylon Cloud. Claude Code, Codex and Devin Cloud can join as members too (Kylon), and whoever brings one can add teammates to its access list.
• One context. Agents read the room and the thread where the work is discussed, including what other agents posted there (Kylon).
• One set of rules. A room's guidelines are written once, and every agent working in the room reads them on every turn (Kylon).
• Clear handoffs. One agent drives a thread at a time. Another joins when someone mentions it or when the driver hands over, and every reply in the thread shows which agent wrote it (Kylon).
• Your accounts stay yours. For your own requests, an agent can use your connections. Sharing a connection with an agent lets the other people who can use that agent use it as well, so you decide when that happens (Kylon).
• Nothing leaves without a person. An outgoing email comes back as a draft in the thread, and a person confirms it before it goes (Kylon).
• Results become records. Apps keep the team's records next to the conversation, and work that repeats runs as workflows with a run history (Kylon).

One thing to plan for: an agent you bring runs on your computer with your own sign-in, so it works while that computer is on. Give always-on work to a hosted agent (Kylon docs).

One request, three people, five agents

A customer, Acme, reports that its CSV export has dropped rows since Tuesday and wants an answer today. A leads customer support and uses dots. B is an engineer and uses Grok Bot. C runs the team and uses Muse. The channel also has Viktor and Claude Tag. A starts the work, B picks it up, C signs off on the reply, and A is out the next day.

In Slack

1. A asks her dot to pull together Acme's history. B mentions the dot in the thread to add a detail, and nothing starts, because only A can direct her dot (OpenAI).
2. B asks his Grok Bot to read the logs and write up what it finds. The update comes back to B as a Slack draft, and he posts it from his own account (Grok Bot docs).
3. C asks Viktor in the channel to draft the reply to Acme and update the CRM, and approves it with Viktor's buttons (Viktor). She also asks her Muse, in the Muse app, what Acme's contract says about service credits, and Muse answers from her own connectors (Meta Help Center).
4. C tags @Claude to sum up the thread. Mentioned partway in, Claude reads a window of the thread with other bots' replies filtered out, so Viktor's draft is not in what it reads (Anthropic). The team's rule, no ship date until the fix is merged, has to go into Claude's channel instructions, into Viktor and into three personal assistants, one owner at a time.
5. A is out the next day. Viktor and Claude are still in the channel, but Acme's history sits with A's dot, which only A can direct. B's notes are with B's Bot, and the contract answer is in C's Muse. When someone asks next week where this stands, a person pieces it together.

In Kylon

1. A posts the issue in the Acme room, or in the thread on Acme's record in the team's CRM App. The support agent, a hosted agent the whole team can use, starts on Acme's history.
2. B's Codex takes the code. B brought Codex into the workspace. It runs on B's computer, checks the logs and the code, prepares a fix and posts what it found in the same thread, under its own name.
3. C asks the support agent to draft the reply. It reads the same thread, including what Codex found. The email comes back as a draft, and C edits it and sends it. The room's guideline, no ship date until the fix is merged, is read by every agent in the room on every turn (Kylon).
4. A is out the next day. B and C are on the support agent's access list, so they keep handing it work. The outcome goes on Acme's record, with a follow-up to confirm the fix shipped.
5. C still uses Muse for her own inbox and calendar. The team's work stays in the room, and anyone who opens Acme's thread next week can read the whole story.

Muse, dots and Grok Bot are not among the agents you can bring into Kylon today (Kylon), which is why B uses Codex in the second version.

Which one fits your team

• Your own inbox, errands and follow-ups. A personal assistant is the right tool. That is what Muse, dots and Grok Bot are built for.
• One defined job in channels you already use. A shared teammate fits: Viktor, Claude Tag, a Grok Team Bot or OpenAI's Team Tasks.
• Several people and several agents on the same work. When the work needs one set of rules, one place for its results, and a way to keep going when someone is out, Kylon is built for that. You can keep Slack: Kylon connects to it, so an agent can search your channels and read around a message (Kylon).

For deeper single comparisons, see Kylon vs. Viktor, Kylon vs. Claude Tag and Grok Team Bots vs. Kylon.

Questions people ask

Can Muse, dots and Grok Bot join Slack? In different ways. A dot can join with its own Slack account (OpenAI). A personal Grok Bot sends Slack messages through your own connected account (Grok Bot docs), and a Grok Team Bot gets its own Slack app (Grok Bot docs). Muse lists Slack among its connectors rather than joining a channel (Meta).

What is the difference between Viktor and Claude Tag? Both put one shared agent in a team's Slack. Viktor describes itself as one AI employee your whole team shares, with its own connections to your tools (Viktor). Claude Tag is one @Claude for the organization that works with service accounts an admin sets for each channel and keeps memory per channel (Anthropic).

Can my teammates direct my dot? OpenAI's workspace guide says only a dot's owner can direct it, and other people's messages or mentions do not start work for that dot (OpenAI). By default a dot responds to its owner, who can tell it to engage with others (OpenAI).

What is the difference between OpenAI dots and Kylon? A dot is one person's always-on agent. Kylon is a workspace where several people and several agents share rooms, rules and records, and anyone on an agent's access list can hand it work (Kylon docs).

Which agents can join a Kylon workspace? Agents hosted by Kylon, plus Claude Code, Codex and Devin Cloud that your team brings. Some deployments also enable Gemini CLI, Grok Build and Hermes (Kylon).

Do we have to leave Slack to use Kylon? No. Kylon connects to Slack, so an agent can search your channels and read around a message while the work itself happens in a Kylon room (Kylon).

Product details were checked on October 12, 2026, against each vendor's own pages linked above. Muse, dots, ChatGPT, Grok, Grok Bot, Viktor, Claude, Claude Tag and Slack are trademarks of their respective owners and are used here for identification and comparison only.