Five products now put an AI agent inside the channel where a team already talks: Claude Tag, Dust, Grok Bot, Runbear, and Viktor. From the outside they look interchangeable. They are not. Two structural choices separate them, and both are decided before you write a single prompt: how many agents you get, and whether the agent lives inside your chat tool or in its own app with chat wired in as a trigger.
The five are listed alphabetically below, not ranked. One disclosure before you read further: Runbear publishes this blog and Runbear is one of the five. Its entry follows the same template as the other four, cons included.
What is an AI agent for Slack?
An AI agent for Slack is software that joins a channel as a bot user, reads the messages it is mentioned in, and acts on connected tools on the team's behalf. It differs from a personal chatbot in one important way: it is shared. Several people talk to the same agent in the same thread, everyone sees the reply, and the next person can pick up where the last one stopped. That shared quality is what makes the identity and permission questions below matter. The category consolidated quickly during 2026. Anthropic launched Claude Tag in beta on June 23, 2026, and retired the older Claude app in Slack on August 3. xAI opened the Grok Bot beta on August 11. Dust raised a $40M Series B in May 2026 and reports more than 3,000 teams.
The two choices that separate these five tools
The first choice is how many agents a workspace gets. Claude Tag and Viktor give you one shared assistant that everyone talks to. Dust, Grok Bot, and Runbear give you several, each scoped to a team or a job. The second choice is where the agent actually lives. Claude Tag, Runbear, and Viktor run inside the chat tool. Grok Bot runs in xAI's own apps and treats a Slack message as one trigger among many. Dust sits between the two, with a web app at the center and a Slack app as one of several surfaces.
Plotting the five on those two axes puts each in a different place. Position reflects a product decision, not quality.
The five at a glance
Tool
Where you ask
How many agents
Who the agent acts as
You pay by
Claude Tag
Slack only
One shared
The workspace
Seat (Claude Team or Enterprise)
Dust
Web, Slack, browser, CLI, API
Many
The workspace
Seat plus monthly credits
Grok Bot
xAI desktop and iOS; chat as trigger
Many
You alone
Bundled in SuperGrok or Cursor plan
Runbear
Slack, Teams, Discord
Many
Each member (Business and up)
Usage credits
Viktor
Slack, Teams
One shared
The workspace
Usage credits
The "who the agent acts as" column is the one most teams overlook. When a shared agent queries a CRM, it either uses one workspace connection for everyone or it uses the credentials of the person asking. That decision determines whether a sales agent can accidentally surface a deal a junior rep should not see.
1. Claude Tag
Claude Tag is Anthropic's Slack-native agent, launched in beta on June 23, 2026, and replacing the earlier Claude app in Slack on August 3, 2026. One @Claude serves an entire channel, so anyone can tag it, watch what it is working on, and continue a colleague's thread. It builds persistent memory of the channels it sits in, breaks a request into stages and works asynchronously, can schedule future work, and in optional ambient mode will surface relevant information without being asked. It is available on Claude Team and Enterprise plans, not on individual Pro or Max.
Pros
Multiplayer by design: one Claude per channel, with shared visibility into what it is doing.
Channel memory means less re-explaining across days and across people.
Admin controls are concrete: token spend limits set per channel and org-wide, plus a log of every action and who asked for it.
No new vendor if you already pay for Claude Team or Enterprise.
Cons
Slack only. There is no Microsoft Teams version, so mixed-channel companies are out.
Not available on Pro or Max, so individuals and small teams on those plans cannot use it.
One shared identity with org-level tool connections. It does not distinguish what each person is allowed to see.
Per-seat pricing means cost scales with headcount, including people who rarely use it.
Ambient mode acting unprompted under one shared identity raises oversight questions some security teams will want answered first.
Best for: teams already paying for Claude Team or Enterprise whose work happens entirely in Slack.
2. Dust
Dust is a Paris-based agent platform where a workspace builds many purpose-built agents, each with its own instructions, base model, scoped knowledge sources, and tool list. Agents are reachable from a web app, a browser extension, a Slack app, a command line tool, and an API. Dust connects to Slack, Google Drive, Notion, Confluence, Microsoft 365, GitHub, Intercom, Salesforce, and Zendesk among others, then continuously indexes them into a per-workspace knowledge base. It raised a $40M Series B in May 2026 led by Abstract and Sequoia and reports more than 3,000 teams. The backend is MIT-licensed on GitHub, so teams with engineering capacity can self-host.
Pros
Many scoped agents, each grounded in a specific slice of company knowledge rather than everything.
Model choice per agent: Claude, GPT, Gemini, or Mistral.
The strongest document grounding of the five, with continuous indexing across a long connector list.
Open-source backend and a self-host path, which matters for teams with data residency constraints.
Enterprise tier adds SSO, SCIM, and a choice of US or EU hosting.
Cons
Slack is one surface among several, not the center of the product, so the in-channel experience is thinner than in chat-first tools.
Pricing shifted during 2026 from flat per-seat to per-seat plus a monthly credit allowance, and public listings still disagree on the current numbers. Check the vendor page before budgeting.
The self-serve tier caps seats, connectors, and spaces, so larger rollouts move to a sales conversation.
Agents act with workspace-level connections, not each member's own permissions.
Getting value takes deliberate setup. An unconfigured Dust workspace does very little.
Best for: teams whose knowledge is scattered across many document systems and who want agents usable outside chat as well as inside it.
3. Grok Bot
Grok Bot is xAI's multi-agent assistant, in beta since August 11, 2026, in which one person runs several specialized agents at once. Each agent gets its own cloud computer and its own memory and can hand work to another agent automatically. Agents stay synced across desktop and mobile and keep working while your device is off. Slack and Teams messages act as triggers, alongside GitHub events and scheduled times, rather than being where the agent lives. A separate Grok app can answer @mentions inside Slack. Grok Bot is bundled into SuperGrok Plus and Heavy, and Cursor Pro+, Ultra, and Teams plans.
Pros
Genuine multi-agent handoff, where one agent passes work to another without you brokering it.
Persistent cloud compute: work continues overnight and while your laptop is closed.
A 220-plugin catalog, with browser control as a fallback when no plugin fits.
No separate line item if your team already holds SuperGrok or Cursor subscriptions.
Cons
The agents live in xAI's apps. If your goal is an agent your team talks to in a channel, this is the wrong shape.
Personal rather than shared. There is no team-visible agent identity that colleagues can tag and follow.
The plugin catalog had no HR or ITSM systems as of August 12, 2026: no Workday, BambooHR, Rippling, Greenhouse, Lever, ServiceNow, or PagerDuty.
Early beta, and availability is tied to specific consumer subscription tiers rather than a company plan.
Best for: individuals who want an always-on personal agent fleet and treat Slack as one input among many.
4. Runbear
Runbear is a channel-native agent platform where a workspace runs several named agents in Slack, Microsoft Teams, or Discord, each with its own knowledge, tools, and instructions. Its distinguishing choice is per-user authentication on a shared agent: on the Business plan and above, one agent can act with each member's own credentials, so a question about a deal returns only the records that person is allowed to see. Pricing is usage credits rather than seats, at $99 per month for Team (20,000 credits, 5 builders) and $399 per month for Business (100,000 credits, 10 builders). Business adds SAML and SSO, an audit log of every prompt, tool call, and reply, and the option to bring your own model key.
Disclosure: Runbear publishes this blog. The pros and cons below follow the same template used for the other four tools.
Pros
Several named agents per workspace, so an HR agent and a platform agent can hold different knowledge and different tools.
Per-user authentication on a shared agent, which none of the other four offer.
Usage-based credits rather than seats, so occasional users do not add cost.
Slack, Microsoft Teams, and Discord, rather than Slack alone.
Scheduled runs and channel triggers let an agent start work without being asked.
Cons
Chat-first, so there is no rich standalone web workspace of the kind Dust offers for long research sessions.
Credit forecasting takes a month or two to calibrate, because a deep research task can cost hundreds of times a simple lookup.
Builder seats are capped per plan at 5 on Team and 10 on Business, which constrains how many people can configure agents.
Per-user auth, SSO, and the audit log sit on Business at $399 per month, not on the entry tier.
Someone has to own agent setup. Teams that skip that step end up with one generic bot and little adoption.
Best for: teams that want several team-specific agents in chat and care that a shared agent respects each person's own tool permissions.
5. Viktor
Viktor is a single shared AI coworker for Slack and Microsoft Teams, built by ZETA AI, Inc. Instead of one assistant per person, one Viktor works in whichever channels the team invites it to, and it reads only those channels. It has its own cloud computer where it writes and runs code, and it returns finished artifacts rather than text: PDFs, dashboards, commits, and deployed internal web apps with a database and auth. Viktor connects to more than 3,200 tools through one-click OAuth, monitors connected data streams to flag issues before anyone asks, and runs recurring scheduled work. New accounts get $100 in free credits with no card required, and paid usage starts at $50 per month.
Pros
Shared rather than per-person, so the whole team sees the same coworker and the same history.
Ships real deliverables. Building and deploying an internal web app is unusual in this group.
More than 3,200 integrations, the largest catalog of the five.
Runs in both Slack and Microsoft Teams, with proactive monitoring rather than reply-only behavior.
Low barrier to trying it: $100 in free credits and a $50 per month entry point.
Cons
One agent only. There is no way to give HR and engineering separate identities, knowledge, or tool scopes.
Operates at the workspace level. Private mode and role-based access controls are still on the roadmap, per the vendor's own FAQ.
SOC 2 Type 1 only, with Type 2 and ISO 27001 in progress, which some procurement teams will treat as a blocker.
Reviewers report a learning curve, slower responses on complex jobs, and occasional loss of context across channels.
Best for: small teams that want one capable coworker producing finished work, without configuring a fleet of agents.
How to choose between them
Four questions narrow five options to one. Answer them in order, because the first one eliminates the most.
If this is true
Look at
The agent should live in its own app, and chat is one trigger among many
Grok Bot
One shared assistant is enough, and you already pay for Claude Team or Enterprise on Slack
Claude Tag
One shared assistant is enough, and you want finished deliverables built and shipped
Viktor
You need several agents grounded in indexed company documents, used from web and CLI too
Dust
You need several agents in the channel that respect each person's own tool permissions
Runbear
These are not mutually exclusive. Plenty of teams run a personal agent for individual work and a shared agent for the channel, and the two rarely conflict.
Which AI agents work in Microsoft Teams, not just Slack?
Runbear and Viktor both run natively in Microsoft Teams as well as Slack, and Runbear adds Discord. Grok Bot accepts Teams messages as a trigger but does not live there. Claude Tag is Slack only, with no Teams version as of September 2026. Dust reaches Teams content as an indexed data source rather than as a chat surface.
Can a shared AI agent respect each person's individual permissions?
Of these five, only Runbear does this, through per-user authentication on the Business plan and above. Claude Tag, Dust, and Viktor all connect tools at the workspace level, so a shared agent sees the same data regardless of who asks. Grok Bot is personal, so it uses your own connections and nobody else's. This matters most for CRM, HR, and finance systems where row-level access differs by role.
What does an AI agent for Slack cost?
Three pricing shapes exist in this category. Per seat: Claude Tag is included with Claude Team and Enterprise seats. Per seat plus credits: Dust charges a seat price with a monthly credit allowance. Usage credits with no seat charge: Runbear starts at $99 per month and Viktor at $50 per month. Grok Bot is bundled into SuperGrok and Cursor subscription tiers. Seat pricing is predictable but charges for people who rarely use the tool; credit pricing tracks real usage but takes a billing cycle or two to forecast.
Do I need Claude Enterprise to use Claude Tag?
Claude Tag requires a Claude Team or Claude Enterprise plan. It is not available on individual Pro or Max plans, which use the older Claude Code in Slack setup path instead. Claude Tag replaced the previous Claude app in Slack on August 3, 2026.
Is one shared agent better than several scoped agents?
Neither is better in general, and the trade is predictable. One shared agent is faster to start and needs almost no configuration, but it carries a single set of instructions and one tool scope for the whole company. Several scoped agents let an HR agent and an engineering agent hold different knowledge and different permissions, at the cost of someone owning their setup. Small teams usually get more from one agent working well. Companies past roughly 100 people tend to hit the limits of a single shared identity.
How this comparison was put together
Every capability and price in this piece was taken from vendor documentation, vendor pricing pages, and launch announcements, checked on September 3, 2026. The five landing page screenshots were captured the same day at a single 1440 by 900 viewport, so they can be compared like for like.
Three things surfaced during that check and are worth stating plainly. xAI publishes no standalone Grok Bot landing page: x.ai/grok-bot returns a 404, which is why the Grok Bot screenshot above shows grok.com, the consumer chat app, instead. Dust's pricing page is client-rendered and returns no content to a plain fetch, and public listings disagree on its 2026 credit migration, so this piece describes the shape of Dust's pricing and leaves the numbers to Dust. Runbear's and Viktor's prices were read directly off their live pricing pages that day.
This is desk research, not a hands-on bake-off. Nobody here ran the same task through all five products. Treat the capability claims as vendor-stated and verified against primary sources, not as measured performance. The category also moves fast: Claude Tag was two months old at the time of writing and Grok Bot three weeks. Check current pages before deciding.
Runbear publishes this blog and appears as one of the five, with the same pros-and-cons treatment applied to it as to the others.
Dust AI vs Runbear (2026): Internal AI Workspace or Customer-Facing Agents?
Dust and Runbear both connect AI agents to workplace tools. The choice that changes your implementation is whether your first problem is governing AI inside the company or deploying an agent into a customer-facing channel.
