How to put an AI agent in your team's Slack channels

You can put an AI agent in your Slack channels without replacing Slack. Connect Slack to an agent-native workspace as one of your connections, give the agent its own identity, tools and computer, and it works in Slack as you do — reading the channels your account reaches, posting, and running tasks in the background. The agent lives outside the Slack tenant, so its memory and tools are not scoped to what Slack can see.

That is the answer, and it is narrower than the phrase “AI agent for Slack” usually implies. There are two different products behind that query. One is AI that lives inside Slack: Slack’s own AI, and agents other vendors embed in the Slack surface. The other is an agent that lives somewhere else and reaches into Slack as one of the tools it uses. They are not competitors at the same layer, and picking the wrong one is why teams end up with a summariser when they wanted work done.

This page is about the second shape. It is the connection step, what the agent can actually do in a channel, and the limits worth knowing before you rely on it.

What “an AI agent in Slack” actually means

Start with what Slack publishes about itself, because it decides which shape you need.

Slack’s own AI is a reading and answering layer inside chat. On its pricing page, read on 2026-10-27, the Free and Pro plans carry “Basic AI”: AI conversation summaries, AI search, AI daily recaps, AI file summaries, and the Slackbot personal AI agent, which “understands the conversations, files, and projects in Slack” [S5]. Business+ adds “Advanced AI”, including AI workflow generation; Enterprise+ adds enterprise search across connected apps [S5]. That is real, useful work, and it happens inside the chat surface on the context Slack can see.

Slack is also no longer only a reading surface. Its agentic platform embeds partner agents — the ones named for it include Claude, GitHub Copilot, Devin and others — and Slack Code spins up a temporary code channel with a repo view, commits and diffs the team reviews in Slack [S7]. The code is written by those partner agents, and it stays inside the Slack tenant [S7].

The third thing in the market is the one most people mean by “an agent in Slack”: Anthropic’s Claude Tag. It puts a shared @Claude into your Slack channels as a team member, and it executes multi-stage work in the thread rather than summarising it [S8]. Anthropic’s own page, read on 2026-10-27, describes it as reading threads, understanding full context and reacting in real time, and states it is “available for Claude Enterprise and Team customers in Slack” [S6].

So the honest framing is not “Slack has no agents”. Slack has AI inside chat, and it now hosts other vendors’ agents inside the same surface. What none of those does is give the agent a home of its own — an identity you administer, tools and memory that are not scoped to the Slack tenant, and a computer to run on. That is the gap the second shape fills.

Connect Slack to your workspace

In Nebula, Slack is a connection, not a host. The pricing page lists it plainly: “Connect your own Gmail, Slack, GitHub, Notion and thousands of other apps. Each connection is yours, and agents use it only on your behalf” [S1].

The mechanics matter more than the label, and Nebula’s documentation is specific about them [S2]:

• Connections are private and personal. They belong to you, not to an agent, and agents use them “only when you ask them to act” [S2].
• Connections travel with you. Connect an app once and a Nebula agent you hand work to acts as you on it, whether you built the agent or a teammate did. When a teammate hands that same agent work, it acts as them [S2].
• You rarely set anything up in advance. When a task needs an app you have not connected, a card appears in the chat and the task picks up again once you connect it [S2].
• There is no account picker on an agent. A Nebula agent acts through the toolkits it has, using the accounts of whoever asked for the work [S2].

That last point is the one to internalise. Connecting Slack does not create a Slack bot with a workspace-wide token; it lets agents act as the person who asked. The Slack permission model you already have keeps applying, because the agent is reaching Slack through your account.

Desktop: Settings → Account → Connections → Add connection. Mobile: Settings → Connections [S2].

Give the agent a job, not a chatbot

A connected Slack account is the door. What makes the difference is what is on the other side of it.

In Nebula, each channel has a team: the set of agents Nebula can hand work to in that channel, and the agents that answer when someone tags them there [S3]. An agent can be on the team in one channel and absent from another [S3]. When you tag an agent that is on the team, it answers for itself, in its own name, and the work “runs as its own task, so it’s never handed to a different agent” [S3].

Each agent carries an identity: its own avatar, name and description, the device where it runs and stores files, its access level, and its own model row [S3]. It has its own connected accounts, and it acts as itself across channels, calls, files and tasks. That is the structural difference from an assistant bolted into a chat tool: the agent is a member with a profile you administer, not a feature of the surface it answers in.

Give it a computer and it stops being something that only replies. A cloud device runs “in the cloud, for your workspace”, and adding a cloud device to a channel shares it with everyone in that channel — they can all run work on it there [S4]. A subscribed workspace starts with one small cloud device that comes with the seat; a bigger size draws more Nebula Credits while it runs [S4].

What the agent can do in a Slack channel

Once Slack is connected and the agent is on a channel’s team, three things become possible that a chat-embedded assistant cannot do.

It works in Slack as you do. Because the connection is your account, the agent can read the conversations and files your account can reach and post back into them, on your behalf and only when you ask it to [S1][S2]. If you want the agent to answer your team inside Slack without moving the team, this is the path.

It runs work in the background. A mention becomes a task the agent owns, and Nebula’s agents pursue multi-step work and report back when they are done or stuck. The work does not have to happen in the thread: it happens on the agent’s device, and the channel gets the result. That is the difference between an agent that answers a question and an agent that does a job.

It hands work on without you relaying it. An agent on a channel’s team can pass part of its work to another one by tagging it in its answer; the tagged agent starts on it as its own task in the same conversation [S3]. So a request that arrives in Slack can fan out to a research agent, a writing agent and a coding agent without a person copying messages between tools.

One shape this makes possible, using only documented behaviour: a customer report lands in a Slack channel, your agent reads it as you, opens a task, does the first pass on the agent’s cloud device, and tags a coding agent on the same channel’s team to take the code change. The human reviews and merges. Nothing about the review changes because the work started in Slack.

Permissions and limits worth knowing

Four limits decide whether this fits, and all four are Nebula’s own.

Personal connections are private. Your Slack connection serves you; a teammate’s connection serves them [S2]. If you want a shared setup, that is a different object — a shared provider key under Settings → Workspace → API keys — and the documentation is clear about which is which [S2].

An agent on your own computer cannot be shared. Nebula’s docs put it flatly: an agent that runs on your own computer “is yours alone — only you can run it, you can’t share it with the workspace, and nobody can add it to a channel” [S3]. If the point is a team using one agent, the agent has to run on a cloud device.

A private agent is not a channel member. A private agent answers you in every channel you tag it in, but it never joins a channel’s team and nobody else can add it to a channel [S3]. Shared agents are the ones a team can hand work to.

We do not claim parity with Slack on the enterprise axis. Slack’s controls are deep — SAML SSO, SCIM, EMM, native data loss prevention, retention and residency, legal holds, audit logs — and Nebula does not match that breadth, nor the size of a directory with over 2,600 apps [S5][S7]. A team whose risk posture runs through those controls should keep Slack for the chat layer. The connection model is designed to let you: the agent reaches Slack through your account, and your Slack governance still applies underneath.

When AI inside Slack is the better fit

This page would be untrustworthy if it claimed the connection model wins everywhere.

If your team lives in Slack and the ask is “help us keep up with Slack”, Slack’s own AI is the right buy and it is included in the plans you already pay for [S5]. If you want a genuine agent that executes work inside Slack without moving, Claude Tag is the lower-friction path, and it is a real agent rather than a summary bot — it breaks tasks into stages and works through them in the thread [S6][S8]. If you want agents embedded in the Slack surface from a vendor you already trust, Slack’s agentic platform is built for exactly that [S7].

The connection shape is for a narrower problem: the agent’s home, memory and tools should not be scoped to the Slack tenant, and the work should be able to leave chat and run somewhere. The full comparison of that axis is on Nebula vs Slack, and the honest version of the Claude Tag question is on Nebula vs Claude Tag.

Cost

Nebula is $25 per seat per month with a 7-day free trial, and a seat covers unlimited messaging, calls and research plus a cloud computer [S1]. You bring your own Claude, ChatGPT or other agent and it runs on the subscription you already pay for, at no Nebula charge; Nebula’s own agents are paid with Nebula Credits, and the workspace asks you to top up before it overspends [S1].

Slack’s plan applies underneath, because Slack is still the host of your channels: Free at $0 with 90 days of message history and up to 10 apps, Pro at $8.75 per user per month monthly or $7.25 annually, Business+ at $18 monthly or $15 annually, and Enterprise+ at $45 per user per month billed annually [S5]. The two are not like for like — one is a chat seat, the other is a workspace seat with a computer — and a team that wants both pays both.

Start here

Pricing for the seat, then the Connections documentation for the connect step. If the category question is still open, start with what an agent-native workspace is; for the tool-access layer an agent uses, read MCP in a team workspace; and for what happens once several agents hold work in one channel, the multi-agent orchestration patterns. Teams choosing a coding agent to connect should read Nebula for coding agents.

FAQ

Can I add an AI agent to Slack without replacing Slack?

Yes. Connect Slack to an agent-native workspace as one of your connections and the agent acts as you in the channels your account can reach. Slack stays your team’s chat tool; the agent lives in the workspace, with its own identity, tools and computer, and reaches into Slack through your account [S1][S2].

How is this different from Slack’s own AI?

Slack’s native AI summarises, searches, recaps and drafts inside Slack, and it is included in Slack’s plans [S5]. A connected agent executes work: a mention becomes a task the agent owns, it runs on its own computer, and it reports back when done. One is a reading and answering layer in chat; the other is a member that does jobs [S2][S3].

How is this different from Claude Tag?

Claude Tag is a shared @Claude that joins your Slack channels as a team member and executes multi-stage work in the thread; Anthropic makes it available to Claude Enterprise and Team customers in Slack [S6][S8]. It is a real agent working inside Slack. Nebula’s agent works from outside the Slack tenant, with its own identity, its own connected accounts and a cloud computer it keeps, so its memory and tools are not scoped to Slack [S1][S3].

Whose Slack account does the agent use?

Yours. Connections belong to the person asking for the work, not to the agent, so the same agent acts as whoever handed it the task — you when you ask, a teammate when they do [S2]. There is no account picker on a Nebula agent; it uses your connections when you ask it to act [S2].

Can a teammate use the agent I set up?

If it runs on a cloud device and is shared with the workspace, yes — it can join a channel’s team and anyone in that channel can hand it work [S3][S4]. An agent on your own computer is private to you: only you can run it, you cannot share it with the workspace, and nobody can add it to a channel [S3].

Do I need a Nebula agent, or can I connect my own?

Both work. Nebula’s own agents are ready to go and draw on Nebula Credits only for the tasks you give them [S1]. You can also bring an agent you already pay for — Claude, Codex or another — and it runs on your own subscription at no Nebula charge [S1]. Coding agents like Claude Code and Codex have their own setup, covered in MCP in a team workspace.

What does it cost?

Nebula is $25 per seat per month with a 7-day free trial, covering unlimited messaging, calls and research and a cloud computer [S1]. Your Slack plan continues underneath: Free $0, Pro $8.75 monthly or $7.25 annually, Business+ $18 monthly or $15 annually, Enterprise+ $45 annually [S5].

Does Nebula replace Slack’s compliance controls?

No, and it does not claim to. Nebula does not match Slack on enterprise compliance breadth — SAML SSO, SCIM, EMM, data loss prevention, retention and residency, legal holds, audit logs — or on the size of its app directory [S5][S7]. The agent reaches Slack through your account, so your Slack governance keeps applying; a team whose risk posture runs through those controls should keep Slack for the chat layer.

Sources

All read on 2026-10-27.

• S1 — https://www.nebula.gg/pricing: $25 per seat per month, 7-day free trial, unlimited messaging, calls and research, a cloud computer, external agents (Claude, Codex or another) on the subscription you already pay for at no Nebula charge, Nebula agents paid with credits and a top-up prompt before overspend, and connections to your own Gmail, Slack, GitHub, Notion and “thousands of other apps” — “Each connection is yours, and agents use it only on your behalf.”
• S2 — https://docs.nebula.gg/docs/workspace/connections (last updated 2026-10-22): connect apps so agents “take real action as you”; personal connections private and used “only when you ask them to act”; connections travel with you, not with an agent; the connect card that appears when a task needs an app; no account picker on a Nebula agent; desktop Settings → Account → Connections, mobile Settings → Connections; shared provider keys versus personal connections.
• S3 — https://docs.nebula.gg/docs/channels/agents: each channel has its own team; an agent can be on the team in one channel and absent from another; tagging an on-team agent makes it answer “for itself, in its own name” and the work “runs as its own task, so it’s never handed to a different agent”; agent identity (avatar, name, description, device, access, model); an agent passing part of its work to another by tagging it; a private agent never joins a channel’s team and nobody else can add it to a channel; an agent on your own computer “is yours alone — only you can run it, you can’t share it with the workspace, and nobody can add it to a channel”.
• S4 — https://docs.nebula.gg/docs/channels/devices (last updated 2026-10-23): a cloud device runs “in the cloud, for your workspace”; a subscribed workspace starts with one small cloud device that comes with the seat; adding a cloud device to a channel shares it with everyone in that channel; your own computer is never shared that way; device sizes and the Nebula Credits they draw.
• S5 — https://slack.com/pricing: Free $0 with 90 days of message history and up to 10 apps; Pro $8.75 per user per month monthly or $7.25 annually; Business+ $18 monthly or $15 annually; Enterprise+ $45 per user per month billed annually; “over 2,600 apps” in the Slack App Directory; Basic AI on Free and Pro (conversation summaries, AI search, daily recaps, file summaries, Slackbot personal AI agent); Advanced AI on Business+; Enterprise-Grade AI and enterprise search on Enterprise+; the 50% off for three months offer on Pro and Business+.
• S6 — https://claude.com/claude-in-slack: Claude Tag, @Claude in Slack, reads threads and reacts in real time, “Available for Claude Enterprise and Team customers in Slack”, coming soon to Microsoft Teams.
• S7 — https://www.nebula.gg/compare/slack: Slack’s agentic platform and Slack Code (temporary code channel with repo view, commits and diffs reviewed in Slack, written by partner agents and staying inside the Slack tenant); Slack’s enterprise controls (SAML SSO, SCIM, EMM, native DLP, retention and residency, legal holds, audit logs) and the over-2,600-app directory; Nebula does not claim parity on those axes.
• S8 — https://www.nebula.gg/compare/claude-tag: Claude Tag launched 23 June 2026; a shared @Claude that joins Slack channels as a team member, breaks a task into stages and replies in the thread; available on Claude Team and Enterprise plans, with channel work billed by usage to the organisation’s usage balance under a spend limit an Owner sets; work runs in a per-session cloud sandbox.