Which AI agent should you let into your Slack, knowing it’ll read your channels and talk to your team? (Choose badly and you’ve hired a bot that replies-all.) The best agents that work inside Slack earn that access with sane permissions, replies that help, and work you can check afterward.
Here’s the TL;DR:
- What makes an agent work inside Slack?
- How I ranked them
- 1. Lindy: best AI teammate inside Slack
- 2. Slackbot: the pick for teams that want AI with no new vendor
- 3. Viktor: strongest at long jobs it runs on its own cloud computer
- 4. Claude Tag: the pick for one shared Claude the whole channel can see
- 5. ChatGPT workspace agents: built for teams that already build in ChatGPT
- 6. Dust: strongest at many specialized agents grounded in company knowledge
- 7. Agentforce: a fit for companies that already run on Salesforce
- 8. Glean: strongest at permission-aware search across enterprise data
- 9. Runbear: the pick for named, shared bots billed by usage
- Which Slack agent fits your team?
- Before you invite an agent into every channel
- FAQ

Lindy is the best agent that works inside Slack in 2026, because it acts like a teammate: it answers in the thread, keeps editable memory of how your company works, and does the work across 1,500+ tools plus MCP, with Approve buttons in shared threads before it writes.
Slackbot is the built-in default, Viktor is a credit-funded AI employee, and Claude Tag puts one shared Claude in a channel. ChatGPT workspace agents suit OpenAI shops, while Dust is a workbench for agent builders.
Salesforce Agentforce in Slack fits Salesforce-run companies, Glean brings enterprise search into threads, and Runbear offers named bots billed by usage.
What makes an agent work inside Slack?
Three kinds of tools can claim the label, and they behave very differently once they’re sitting in your sidebar.
Built-in agents. Slackbot ships with Slack itself, so there’s no install and no new vendor to review (procurement sends its thanks). What you get depends on your Slack plan tier.
Assistant apps you tag. Claude Tag, ChatGPT workspace agents, Dust, and Glean join channels as apps you @mention for an answer or a task, and each brings its own models and connected data along.
AI coworkers that take on work. These pick up a task from a thread, run it across your tools, remember what the team decided last time, and come back with something finished. Viktor and Runbear sit here, and so does the agent at the top of this list.
How I ranked them
Where it replies. The thread you asked from beats a DM you have to go find, because the thread is where everyone else can follow along.
What it can act on. Reading and summarizing is table stakes, so agents that can open, update, and send things in connected tools ranked higher, especially with a human gate on the writes that matter.
What it remembers. An agent that forgets last week’s decision makes you repeat yourself, so memory you can see and correct counted for a lot.
How billing works. Per seat, per credit, per conversation, or bundled into a plan you already pay for, since the model decides how the bill grows once the whole team starts tagging it.
1. Lindy: best AI teammate inside Slack
The top spot goes to the agent that behaves most like a colleague who’s been around a while, because it already knows the context when you tag it.
Who should add it: Slack-based teams that want one AI teammate anyone can @mention, whether that’s sales chasing a follow-up or ops pulling a weekly report. An admin installs it in minutes with no API keys or IT ticket, and everyone else links with one tap the first time they message it.
How it works in Slack: An @mention in a channel gets a reply inside that thread, never as a fresh top-level post, and after that first reply you can keep talking without tagging it again.
A shared channel only gets its public-safe tools, meaning the web, public Slack search, and Notion when it’s connected, so calendar, email, and reminders wait for your DMs and the threads you started.
Company knowledge sits in plain memory files that anyone can read and correct, and recorded calls go into a shared meeting library, which spares the rest of the team from sitting in on every meeting.
It reaches 1,500+ tools plus any MCP server, and where no integration exists it can drive the browser the way a person would.
Each integration gets a guardrail: Always allow, Require approval, or Don’t offer. In a shared thread, a write set to Require approval shows up as Approve and Deny buttons that the requester or a workspace admin can click.
Pricing: Free is $0 with $50 in credits for 7 days and no card. On Team, you pay $49.99 a month for each active user, every user brings 5,000 credits into one shared pool, and extra credits come at $10 per 1,000. Enterprise is custom and adds SSO, SCIM, and audit logs.
Limitations: It’s Slack-first, so a company that lives in Microsoft Teams is a weaker fit, since Teams chat isn’t supported. New installs also need a paid Slack plan.
2. Slackbot: the pick for teams that want AI with no new vendor
Slackbot is the easiest agent here to say yes to, because it’s already in your sidebar and there’s no extra vendor for security to vet.
Who should add it: Teams already on Slack Business+ or Enterprise+ that want an AI agent without signing another contract. Owners and admins can switch it on or off for the whole workspace or org, which keeps the rollout simple.
How it works in Slack: There’s nothing to install. You tag Slackbot into a channel to pull a long thread together and share the recap back, and it searches your Slack history, connected apps, and the web, then returns reports with citations.
Slack says it can act across apps such as Jira, Figma, Salesforce, and Docusign through MCP. You can also build repeatable workflows by asking, pin dashboards, polls, or calculators to a channel, share skills, and use Slack CRM to create accounts, update records, and log calls.
On privacy, Slack’s line is that it surfaces only what you already have access to, keeps your interactions private, and never uses your data to train AI. There’s voice input on mobile, too.
Pricing: Slackbot comes with your Slack plan. As of October 2026, Slack Pro is $8.75 per user per month and Business+ is $18 per user per month, with Enterprise+ through sales, and Free and Pro workspaces only get Slackbot during a trial or preview.
Limitations: Each Business+ member can send Slackbot 15 messages a week, which a heavy user could spend before the first coffee wears off. The full agent also means paying Business+ for everyone on the workspace.
Enterprise plans are limited only by Slack’s overall usage policies, and Salesforce Flex Credits buy extra usage when a team outgrows its allowance.
3. Viktor: strongest at long jobs it runs on its own cloud computer
Who should add it: Teams that like paying for one workspace-wide credit pool and want to hand off long jobs like reports, dashboards, code, and campaigns. It also works in Microsoft Teams, which helps if your company is split between the two.
How it works in Slack: Mention @Viktor in any channel or DM, and it picks up the task, runs it on its own cloud computer, and replies with the finished result.
A single Slack message can send it across Stripe, Notion, HubSpot, GitHub, Linear, and Google Drive, and it keeps your team’s decisions and preferences between jobs.
It asks before anything it can’t undo, and it can run work on a schedule. Viktor lists 9,800+ integrations and Viktor Spaces, describes itself as SOC 2 compliant, and says it passed Slack’s security review.
Pricing: Free costs $0 with up to $100 in credits and no credit card as of October 2026, and those free credits never expire. Team is $100 a month for 40,000 credits shared by the whole workspace with no per-seat charge, unused monthly credits roll over for one month, and Enterprise is custom.
Viktor’s own guide puts a quick Slack question at 50 to 100 credits and a deep report at 500 to 2,000, so 40,000 credits covers roughly 80 to 400 tasks. Top-ups and auto-top-ups with a monthly spend cap are available, and nonprofits get 15% off.
Limitations: That spread makes monthly spend hard to predict until you’ve watched a few weeks of usage. Scheduled automations draw from the same pool, so a long list of recurring jobs eats credits faster than you’d guess.
4. Claude Tag: the pick for one shared Claude the whole channel can see
Who should add it: Organizations already on Claude Team or Enterprise that want Claude working in the open with everyone. Anthropic launched it in beta on June 23, 2026, as the replacement for its earlier Slack app.
How it works in Slack: Admins grant access to selected channels and connect tools, data, and codebases. Anyone in the channel then tags @Claude to delegate, and because one Claude interacts with everyone there, the whole room can see what it’s working on.
It builds context by remembering relevant information from its channels, and that memory stays scoped, so a sales setup and an engineering setup don’t share what they’ve learned.
An optional ambient mode flags useful information and follows up on stalled threads, and it can work asynchronously and schedule its own tasks.
DMs respond privately using your personal tools and connectors on your own claude.ai account. Channel work bills to the organization, and by default nobody needs a Claude account to tag it in a channel.
Admins get token-spend limits for the org and per channel, a log of everything it did and who asked, and spend analytics by channel. Each channel can also pick its model.
Pricing: There’s no separate Claude Tag price. As of October 2026 it needs Claude Team, where a Standard seat is $25 per month and a Premium seat is $125 (for teams of 2 to 150), or Enterprise, with channel work billed to the org as usage.
Limitations: It’s still a beta gated to Team and Enterprise plans. Because channel usage is billed on top of seats, cost climbs with how often your team tags it, and it doesn’t report from private channels, so some of your busiest conversations stay out of reach.
5. ChatGPT workspace agents: built for teams that already build in ChatGPT
Who should add it: Companies on ChatGPT Business, Edu, or Enterprise that want the agents they’ve built in ChatGPT to show up where the team talks. Personal ChatGPT accounts can’t use workspace agents.
How it works in Slack: The ChatGPT Agents app brings workspace agents into Slack, and each agent gets its own Slack handle, backed by a Slack user group. For every channel you choose whether it responds to all relevant messages or only replies when @mentioned, and it can follow up in the thread.
It works in private channels once the app is added, can be scheduled to post results to a channel, and can send emails and update documents without anyone leaving Slack. Search respects Slack permissions, so it only sees the messages and files you could already open.
On the ChatGPT side, Business connects Google Workspace, Slack, GitHub, and Microsoft 365, adds usage analytics and spend controls, and doesn’t train on business data by default.
Pricing: ChatGPT Business Standard is $25 per user per month as of October 2026, with Premium at $125 and a 2-user minimum. You’ll also need a paid Slack plan.
Limitations: Setup has three parts (the ChatGPT app, the Slack app, and permission for members to create user groups), plus extra org-level steps on Enterprise Grid. OpenAI also warns that ChatGPT Agents’ responses may be inaccurate, so read anything important before it goes out.
6. Dust: strongest at many specialized agents grounded in company knowledge
Dust is the pick when you want a shelf of purpose-built agents (one for support, one for sales, one for ops) more than a single generalist.
Who should add it: Anyone planning to build lots of specialized agents, ideally with a teammate who enjoys designing them.
How it works in Slack: An admin turns on the Slack Bot under Admin > Integrations. Members then mention @Dust in a channel or thread, or DM the app with no mention needed, and the agent replies in a thread.
You choose which agent answers with +AgentName, one agent per message, and admins can set a default agent per channel or let it auto-reply without an @mention. Agents can also run from Slack Workflow Builder.
Behind those agents sit 20+ models (GPT, Claude, Gemini, Mistral, and DeepSeek), custom skills, knowledge, and tools, multi-agent workflows on schedules and triggers, MCP servers, and dozens of connectors. Frames build interactive dashboards and apps, and Pods give teams shared workspaces.
Security covers SOC 2 Type II, SSO through Okta, Entra ID, or Jumpcloud, and US or EU data residency.
Pricing: Dust lists prices in euros, excluding VAT. As of October 2026, Free is €0 with 500 lifetime credits, Pro is €30 per seat per month with 8,000 credits, Max is €150 per seat per month with 40,000 credits, and Enterprise is custom.
Limitations: Being builder-first, it needs someone to design and maintain the agents. Credits reset monthly with no rollover, and US teams get to do currency math on every invoice.
7. Agentforce: a fit for companies that already run on Salesforce
Who should add it: Salesforce customers that want agents built around their own data and permissions. Slack’s help center says it’s available on all Slack plans with a Salesforce Agentforce license.
How it works in Slack: People at your company create each agent for a specific purpose, such as answering IT help requests or explaining benefits. You find them under Tools > Agents and DM them with prompt starters and past threads, or add one to a channel and @mention it like any teammate.
Agent prompt links can be shared in channels, canvases, and workflows, and admins decide who gets access. Agents use your Slack and Salesforce permissions and won’t reply with information you can’t see.
Department agents cover IT, HR, sales, and service, and your builders work in Agentforce Builder, Prompt Builder, and Agent Script.
Pricing: As of October 2026, Flex Credits cost $500 per 100,000 credits, customer-facing conversations are $2 each, and the Agentforce User License is $5 per user per month (Flex Credits required).
Add-ons run $125 per user per month ($150 for Industries), and Max Edition starts at $550 per user per month. Salesforce’s rate card still sends you to a sales rep for detailed pricing, and “Try for free” appears beside the builder tools.
Limitations: Your company has to build and configure the agents, and admins assign access, so day one starts with a build project. Most of the value also depends on holding an Agentforce license and keeping your data in Salesforce.
8. Glean: strongest at permission-aware search across enterprise data
Who should add it: Larger companies with knowledge spread across many systems that want answers respecting each person’s permissions. Glean pitches itself as enterprise AI that understands your company.
How it works in Slack: The Slack app finds what you need and recaps key conversations without leaving Slack. It can intercept questions that would otherwise land on IT, HR, or engineering, and it stitches together data from Slack, Jira, GitHub, and Confluence when someone’s debugging.
The app asks for permission to view channel content and to perform actions as you and in channels. It runs on OpenAI GPT, Google Gemini, and Claude models with 0-day LLM retention, and the Assistant view in Slack needs a paid Slack plan.
Beyond Slack, Glean offers more than 250 connectors, Glean Agents from its October 2026 Drop, and a pitch for deploying a governed AI coworker with enterprise controls and permissions.
Pricing: Glean doesn’t publish prices as of October 2026. Its pricing page redirects to the homepage, the main button asks you to get a demo, and there’s no self-serve trial, so plan on a custom quote.
Limitations: Everything runs through sales, which means you can’t kick the tires on your own. It also pays off once many of your sources are connected, so a small team with three tools and a shared drive won’t see much of what it can do.
9. Runbear: the pick for named, shared bots billed by usage
Who should add it: Anyone who wants several named agents, like @SalesOps, @Support, or @HR, each with an owner and only the tools you give it. It also covers Discord and Microsoft Teams, handy for a company spread across chat apps.
How it works in Slack: Anyone in the channel can @mention an agent, and it answers in the same thread, tagging a human when it isn’t sure. Meeting recaps and drafts land in your DMs, and agents can trigger on keywords.
Runbear connects to 2,000+ tools with MCP actions, supports personal agents and multi-agent selection, and handles per-user auth on shared agents. A SOC 2 Type II report is available on request.
Business adds an audit log of every prompt, tool call, and reply, plus SAML and SSO. Business+ lets you bring your own LLM key, and eligible Enterprise accounts can get HIPAA with a BAA.
Pricing: Team is $99 a month for 20,000 credits and 5 builders as of October 2026, Business is $399 a month for 100,000 credits and 10 builders, and Enterprise is custom. Yearly billing saves 20%, Runbear charges by usage with no seat fee, and it’s free to start with no credit card.
Limitations: Monthly credits reset, and agents pause when the pool runs dry unless pay-as-you-go is on. Only 5 people can build agents on Team, which can bottleneck a busy workspace.
Which Slack agent fits your team?
Start with what you already pay for, because three of these ride on contracts you may have signed years ago.
If you’re on Business+ and want zero procurement, give Slackbot a real trial and see whether 15 messages a week per person covers your heaviest users. Salesforce shops with admins who like building should look hard at Agentforce.
If you standardized on ChatGPT Business or Claude Team, the matching agent is the shortest path, since the seats and admin controls are in place. Builders who want a shelf of specialists belong on Dust, while Glean suits buyers whose bigger problem is finding things across a dozen systems.
Viktor and Runbear make sense for teams that prefer one usage-based bill. And when the goal is one teammate that takes a task from a thread to done, with memory you can edit and approvals on the writes, the #1 pick is where I’d start.
Before you invite an agent into every channel
Every agent on this list asks for some level of access, and the defaults deserve a careful look before the whole company starts tagging.
Start in one channel. Pick a busy internal room with forgiving stakes and let the agent prove itself there before you add it anywhere customers come up.
Check private channels and DMs. Some agents work in private channels once they’re added, others won’t report from them, and DMs often run on personal connections, so read each vendor’s rules before you assume anything.
Decide who approves. If an agent can write to your tools, name the people who can approve those writes in shared threads, and save “always allow” for actions you’d be comfortable seeing happen without a check.
FAQ
Which is the best AI coworker for Slack?
Lindy is the best AI coworker for Slack for most teams, because it replies in the thread, keeps editable memory of how your company works, and can ask for approval before writing in a shared thread. Viktor is the closest alternative if you’d prefer one shared credit pool to per-seat billing.
How much does Slackbot cost?
Slackbot comes bundled with Slack, but the full agent starts at Business+, which costs $18 per user per month as of October 2026. Free and Pro workspaces get it only during a trial or preview, and each Business+ member can send it 15 messages a week.
Do Slack AI agents need a paid Slack plan?
Most of them do, with Agentforce as the main exception, since Slack lists it on every plan once you hold a Salesforce Agentforce license. ChatGPT workspace agents and Glean’s Assistant view both require a paid Slack plan, and Slackbot’s full agent arrives only with Business+ or Enterprise+.
