The fastest way to add an AI agent to a Slack channel without engineers is a hosted no-code agent builder: you describe the job, connect the documents and tools it needs, install its bot to your workspace, and invite it to the channel. Nothing runs on your own server. In Runbear this takes one sitting, and the agent answers in Slack the moment you mention it.
Three ways to add an AI agent to Slack
1. Build your own Slack app with Bolt and an LLM API
Slack's Bolt framework (JavaScript, Python or Java) plus an API key from OpenAI, Anthropic or Google gives you complete control. What it takes:
- An app manifest and scopes. You create the app at api.slack.com, request bot scopes such as
app_mentions:read,chat:writeandchannels:history, and subscribe to events. - Hosting. Slack has to reach your code over the Events API, or your code holds a Socket Mode connection open. Either way, something has to run 24/7, and with the Events API it has to acknowledge each event within three seconds or Slack retries it.
- Everything around the model call. Thread context, retrieval over your documents, keeping them in sync, and storing OAuth tokens if the bot acts in other tools.
- Maintenance. Model versions retire, SDKs change, and someone gets paged when the bot goes quiet.
Choose this if a developer will own it and you need behavior no product offers. For an ops manager working alone, it is a software project to own and maintain.
2. Native assistants: Slackbot, Claude and ChatGPT in Slack
As of October 2026:
- Slackbot (Slack AI). Slack's pricing page lists Slackbot as a "personal AI agent" on Business+ and Enterprise+. Slack says Free and Pro users will soon get a limited trial.
- Claude in Slack. Anthropic's Claude Tag, in public beta, replaces the earlier Claude app in Slack. It puts @Claude in channels for Claude Team and Enterprise organizations and bills channel work by usage against a spend limit. Individual Pro and Max accounts can still use the earlier app, mainly for Claude Code, in workspaces not connected to Claude Tag.
- ChatGPT for Slack. The Slack Marketplace listing says the app is available to ChatGPT Business and Enterprise subscribers.
If your company already pays for one of these plans, start there. They are quick to turn on and maintained by the vendor. The trade-offs are that each one ties you to that vendor's models and plan tiers, and you get the vendor's choices about per-channel behavior, which documents it can read, and which tools it can act in.
3. A no-code agent builder like Runbear
A no-code AI agent for Slack sits between the two. Runbear hosts the agent, keeps your knowledge sources synced, and runs the managed tool integrations, so there is no server to keep alive. You choose the model, the sources, the tools and the reply behavior per channel. It costs a subscription and less control than your own code, and it works on any Slack plan your workspace lets you install apps on.
How to add an AI agent to Slack with Runbear
The example is an ops agent for #ops-help that answers from your runbooks and files Linear tickets.
Step 1: Describe the job
Open Agents and select Create Agent. Name it for its job, such as Ops Helper, and keep the Standard option, which usually creates a Claude Agent SDK agent. (Runtimes)
Open the Instructions tab. Write the core job in Main instructions, and use Add section for separate topics such as escalation rules or answer format. A useful first draft says who the agent serves, where its answers come from, and what it should do when it does not know:
You answer operations questions for the support and fulfillment teams using our runbooks. Quote the runbook section you used. If the runbooks do not cover the question, say so and suggest posting in #ops-leads instead of guessing.
Step 2: Pick a model
Open Settings and go to Model Selection. Claude Agent SDK agents offer Claude, GPT and Gemini models. For an agent that mostly answers from documents, Runbear's own guidance is that the cheapest tier usually answers as well as the most expensive one and uses a fraction of the credits. Save the larger models for multi-step analysis. (Agent settings)
Step 3: Connect knowledge
Open the Knowledge tab and pick a source: Notion, Google Drive, Confluence, SharePoint, Slack, Website or Local Files. Authorize the account, choose the pages, folders or URLs to include, and Runbear runs the first sync. Later syncs are incremental, so edits to your runbooks reach the agent without a manual refresh. (Knowledge overview)
A website source crawls a URL and the subpages under it, up to 200, and re-crawls daily. (Website)
Step 4: Connect tools and actions
Knowledge answers "what do our docs say." Tools answer "what is true right now" and "do this for me." Open the Tools tab, pick a service from the managed catalog (Linear, Jira, HubSpot, Salesforce, Notion, Airtable and others), and choose how people authorize it, where the tool supports both options:
- Per-User: each member connects their own account, and the agent acts with that member's permissions.
- Shared: one account serves everyone who uses the agent.
For internal systems, a Custom MCP server lets you connect your own API. (Tools overview, Tool catalog)
Step 5: Install to Slack and invite it to a channel
Open the Chatbots tab and click the Slack tile. Choose how the agent appears:
- Create @Ops Helper (Create Custom Bot, recommended) gives the agent its own name and avatar. It asks for a Custom App Name and has you select a Slack Configuration Token, and the form links to instructions for generating the token.
- Use @Runbear uses the shared Runbear bot. If it is not installed yet, click Install @Runbear to Slack and approve it in Slack.
Under Create Connection, pick #ops-help (or Direct Message with Runbear for DMs) and click Create. Then invite the bot to the channel in Slack. You need permission to install apps in your workspace, or an admin who will approve the request. (Slack setup)
Step 6: Choose reply behavior
Each channel connection has its own Trigger tab, so the same agent can behave differently in different channels. (Chatbot settings)
- Always replies to @mentions and DMs is always on.
- Continue in threads (after first reply) lets the agent keep answering in a thread it has joined without another @mention, when it has something useful to add.
- Reply without @mention answers every new message in the channel. It is useful in a dedicated help channel and noisy anywhere else.
- Trigger by keyword answers messages that contain words you choose, such as "refund" or "PTO."
Agents on other runtimes also offer Trigger on a schedule, Trigger by emoji reaction and Trigger by sender, plus a Preferences tab for options like Only visible to selected users and Use full thread as context.
Step 7: Test before anyone sees it
Open Improve & Try and ask the questions your team will ask, including ones you suspect it cannot answer. It uses the agent's real instructions, knowledge and tools, so tool calls run for real: asking it to file a ticket files a ticket. (Testing)
Step 8: Roll it out to the team
Tell the channel what the agent answers and who to ask when it can't. After launch, the Activity tab shows each conversation with the trace behind it. On Slack you can turn on Enable response feedback in Settings to add thumbs-up and thumbs-down buttons under each reply, which tells you quickly which answers need work.
Shared agent or personal assistant?
A personal assistant works for one person, with that person's accounts. A shared agent lives in a channel and works for everyone in it. Shared AI agents for teams are usually the better fit for ops work, because questions and answers stay visible and one person maintains the instructions for everybody.
The hard part of a shared agent is permissions. If its CRM or ticketing tool is connected with one Shared account, everyone who talks to the agent gets that account's access. With Per-User authorization, each person connects their own account, and the agent acts with that person's permissions. Two people asking the same question can get different answers, which is what you want when a manager can see deals a new hire cannot.
Access control in an agent's settings decides who can view and edit the agent in the dashboard. It has no effect on who can talk to it in Slack. Restricting which Slack users can start the agent on a connection, and requiring admin approval for new tools, are both in limited availability; contact support to enable them.
Troubleshooting and limits
- The agent does not reply. Check that the bot is invited to the channel and that the trigger you expect is on. An agent that answers correctly in Improve & Try but not in Slack usually has a channel configuration issue.
- Emoji or sender triggers ask you to reinstall. The Slack installation lacks a permission. Reinstall from Settings → Integrations → Slack; only the person who installed it can do that.
- A setting from this guide is missing. Claude Agent SDK agents deploy to Slack and Microsoft Teams only, have no Preferences tab, and offer four Slack triggers. Schedules, emoji and sender triggers need a Classic, Anthropic or OpenAI Responses agent.
- Answers are confident and wrong. The source is not synced or does not contain the answer. Check the Knowledge tab before rewriting instructions. Google Drive and SharePoint files in formats Runbear cannot index are marked Unsupported.
- The agent stops mid-month. Agents pause when the organization reaches its credit limit, and Runbear emails before that happens. On a paid plan an Admin can set an Overage limit. Unused credits do not roll over.
What it costs
Runbear is usage-based, not per seat. As a rough guide from the docs, a question answered from knowledge uses 10 to 20 credits and a task with several tool calls uses 30 to 50 or more. (Billing and credits)
- Team: $99/month, 20,000 credits, 5 builders.
- Business: $399/month, 100,000 credits, 10 builders, plus an audit log of configuration changes, SAML SSO and bring-your-own-key.
- Enterprise: custom.
See pricing for current details.
FAQ
How can an ops manager set up an AI agent in Slack without an engineering team?
Use a hosted no-code agent builder such as Runbear. Create an agent, write its instructions, connect the documents it should answer from (Notion, Google Drive, Confluence, SharePoint or a website), install its bot to Slack, and invite it to a channel. The only admin step is Slack app approval, if your workspace requires it.
Which AI chatbots work with Slack without any coding?
As of October 2026: Slackbot on Slack Business+ and Enterprise+, Claude Tag on Claude Team and Enterprise, ChatGPT for Slack on ChatGPT Business and Enterprise, and no-code agent builders such as Runbear, which let you choose the model, knowledge sources and tools per agent.
What's the easiest way to host an AI agent for Slack without running my own server?
Use a product that hosts the agent for you. Runbear runs the agent, syncs your knowledge sources, and hosts the managed tool integrations, so you install a Slack bot and never deploy code. Building with Bolt means running a server or Socket Mode process yourself.
How can a team share one AI agent in Slack while each person keeps their own tool permissions?
Connect the agent's tools with per-user authorization. Each person links their own account, and the agent acts with that person's permissions instead of one shared login. In Runbear, you choose Per-User or Shared for each tool you connect.
Does the agent read every message in my Slack workspace?
No. The agent answers only in the channels and DMs you connect it to. It always answers @mentions and DMs there, and answers other messages only if you turn on triggers such as Reply without @mention or Trigger by keyword. Slack becomes a knowledge source only if you add specific channels on the Knowledge tab.