How to add an AI assistant to Slack or Teams

By the end of this guide you will have an AI assistant answering questions inside Slack or Microsoft Teams, drawing on your team's own documents, limited to the people and channels you choose, with a plan for what it may and may not do. The job is done by a workspace AI assistant: a bot or agent that lives in your chat tool and connects to your knowledge sources.

To add an AI assistant to Slack, the main choices are who builds and maintains it, which sources it reads, and how its access is controlled. This guide walks through setup and rollout and compares six options from our Chat assistants category.

Key takeaways

• Pick the assistant type first, then connect the knowledge sources, set permissions, test in a private channel and roll out gradually.
• The main decision is between an agent platform you configure in a dashboard, a knowledge platform that answers from governed company content, and a self-hosted option for engineering teams.
• Costs differ in kind: Runbear and Notion meter AI use through credits, Leo shares a monthly token pool, and Guru is scoped through a sales conversation.
• An assistant can only be as safe as its permissions. Limit what it can read and do, and have a person review actions that affect customers, money or systems.

What you need before you start

• Admin rights. You need permission to add apps to your Slack or Teams workspace, or a colleague who has it.
• Your knowledge sources. Document stores, wikis or help-centre content the assistant should read, and the right to share them.
• A pilot group. A small team or private channel to test with.
• A list of allowed actions. What it may answer, and which actions in other tools, such as creating tickets, are acceptable.
• Time. A basic pilot takes under an hour for a dashboard-based tool. Knowledge clean-up and rollout take longer.
• Difficulty. Moderate for dashboard tools, harder for self-hosted options that need Kubernetes.
• Policy. Check your company's rules on AI, data handling and any regulated information before connecting sources.

Which kind of AI tool fits this job

Pick by who will maintain the assistant and what it needs to know.

Agent platforms configured in a dashboard. Runbear lets a team set up AI agents inside Slack, Microsoft Teams and Discord, with HubSpot and Zendesk channels on higher plans, that answer questions from connected knowledge sources and take actions in other tools. It lets you choose Claude, GPT or Gemini per agent and uses per-user authorization, so connected tools act as the person who made the request. ChatBotKit assembles agents from datasets, skillsets and integrations, and deploys them to Slack, Microsoft Teams, Discord, WhatsApp, web widgets or the API.

Enterprise assistants with role-aware access. Leo gives a team one place to work with several models, connects to company documents in OneDrive and Google Drive, answers according to each person's role, and works inside Microsoft Teams and Slack.

Knowledge platforms that feed answers. Guru pulls company information from more than 100 sources into one governed index, inherits permissions from the original systems and works inside Slack and Microsoft Teams. Notion offers Custom Agents that run recurring work such as answering questions in Slack, on top of its workspace of pages and databases. Notion lists agents from the Business plan up, and Custom Agents are free to try, then use paid credits.

A self-hosted control plane for engineering teams. Agent Barn is an open-source control plane for running agents that you talk to in Slack, Microsoft Teams or Telegram, deployed with Helm to Kubernetes. It suits teams comfortable operating their own infrastructure.

Step 1: Define what the assistant is for

You end this step with a written scope that keeps the pilot focused.

1. Pick two or three jobs, such as answering policy questions, searching documents, or drafting replies from a knowledge base.
2. Decide who it serves first: one team, one channel, or the whole workspace.
3. Write what it must not do, such as answering HR cases or touching customer data.
4. Choose a measure of success, such as fewer repeat questions to a particular person.

Check it worked by having a one-paragraph scope that a colleague can read and agree with.

Step 2: Choose the assistant that fits your team

Match the tool to your people and platform.

1. If a non-engineer will configure it, look at dashboard tools such as Runbear.
2. If you want answers from one governed knowledge base, look at Guru or Leo.
3. If your team already lives in Notion, look at its Custom Agents.
4. If you run your own infrastructure and want control, consider Agent Barn. If developers want to build agents visually or in code, look at ChatBotKit.
5. Confirm it supports your chat tool, since Slack, Microsoft Teams and Discord support differ by product.

Check it worked by confirming the tool lists your chat platform and knowledge sources.

Step 3: Connect your knowledge sources

The assistant is only useful if it can read the right material.

1. Start with a small, clean set of sources, such as one wiki space or a folder of policies.
2. Connect them using the tool's integrations. Runbear lists Google Drive, Notion, Linear and Confluence, Leo lists OneDrive, Google Drive and SharePoint, and Guru has 100+ connectors.
3. Remove outdated or conflicting documents first, because the assistant may repeat both.
4. Check how the tool handles permissions. Guru says permissions are inherited from the original systems, and Leo uses role-based access.

Check it worked by asking a question whose answer sits in one of the connected documents and getting that answer.

Step 4: Add the assistant to your workspace

Install it where your pilot group will see it.

1. Follow the tool's setup flow to add it to Slack or Microsoft Teams, with an admin's approval.
2. Create a private test channel and invite the assistant and your pilot group.
3. Name and describe the assistant so people know what it is for.
4. Choose the model if the tool lets you. Runbear lets you pick Claude, GPT or Gemini per agent.

Check it worked by mentioning the assistant in the test channel and getting a reply.

Step 5: Set permissions, rules and guardrails

This step decides whether the assistant is safe to roll out.

1. Limit which channels and people can use it.
2. Decide what actions it can take. Runbear's per-user authorization means connected tools act as the person asking, which keeps actions within that person's access.
3. Write its instructions: tone, what to answer, when to say "I don't know" and who to contact.
4. Turn on logs or audit features if available. Runbear lists an audit log of prompts, tool calls and replies on Business, and Agent Barn records conversations, tool calls and model costs.

Check it worked by asking it to do something it should refuse and seeing it decline.

Step 6: Test with real questions in a pilot

Real use shows the gaps.

1. Collect ten or so questions the team really asks and run them.
2. Check each answer against the source, including edge cases and out-of-date topics.
3. Fix gaps by improving or adding documents rather than only editing the prompt.
4. Gather quick feedback from the pilot group after a week.

Check it worked by having most test questions answered correctly, with the rest either fixed or deliberately declined.

Step 7: Roll out, watch usage and keep it current

Widen access once the pilot holds up.

1. Announce the assistant, say what it is for, and say what it should not be asked.
2. Start with one more team, then the wider workspace.
3. Watch usage and cost. Runbear agents pause when credits run out, and Leo's plans share a monthly token pool.
4. Review answers and logs every month, and update content when policies change.

Check it worked by seeing steady use and fewer repeat questions after a few weeks.

Tools for this job compared

Tool	Good fit for	Pricing	Stated limits
Runbear	Shared agents in Slack, Teams and Discord, configured in a dashboard	Freemium, paid from $99/mo (checked 24 Oct 2026)	Agents pause when credits run out; monthly credits do not roll over; HIPAA needs an eligible Enterprise agreement
ChatBotKit	Developers building agents deployed to Slack, Teams, Discord and web	Free trial, paid from $25/mo (checked 24 Oct 2026)	Basic includes 5 bots and 5K messages per month; widget branding removal from Pro up
Leo	Role-aware multi-model assistant in Teams and Slack	Pricing unknown	Starter is for up to 10 people on a shared token pool; Advanced has a shared monthly pool; early bird prices coming soon
Guru	Governed company knowledge delivered in Slack and Teams	Contact sales	No public price list; packages scoped through sales; AI work consumes credits from a shared pool
Agent Barn	Self-hosted fleets of agents in Slack, Teams or Telegram	Open source, paid from $20/mo (checked 24 Oct 2026)	Cloud is free for a limited time and may change; model inference billed separately; Cloud accounts by invitation
Notion	Custom Agents inside a Notion workspace, answering in Slack	Freemium, paid from $10/mo (checked 24 Oct 2026)	AI on Free and Plus is limited and meant for trying; Custom Agents free to try, then use paid credits; some features in beta

Prices and plan limits change, so confirm the current terms on each vendor's site before you subscribe.

Common mistakes to avoid

• Connecting everything on day one. A large, messy knowledge base produces confusing answers and exposes more than you intended. Start with a small, clean set and add sources one by one.
• Giving broad access to the assistant. An agent with wide permissions can act beyond what a person should. Prefer tools that act as the requesting user, and limit channels and actions.
• Skipping the pilot. Rolling out untested means colleagues meet wrong answers first. Run real questions in a private channel and fix gaps before you announce it.
• Ignoring credits and limits. Usage-based pricing can run out mid-month, and Runbear agents pause when credits are used up. Estimate usage and set alerts or reviews.
• Treating answers as final. Assistants can be wrong or outdated. Tell people to verify anything that affects customers, money, contracts or compliance with the source or a person.

How we put this together

This is an evidence-based guide built from each tool's listing in our directory, which is compiled from the vendor's own site and pricing page. It is not a review based on using the tools. We don't rank tools, and nobody paid to be included. Some outbound links may be affiliate links; see our disclosure.

Frequently asked questions

Is there a free way to add an AI assistant to Slack?

Partly. Runbear has a free plan where the included credits are the hard limit, ChatBotKit's Basic plan comes with a 7-day trial, and Agent Barn's hosted Cloud is free for a limited time. Notion's AI on Free and Plus is limited and meant for trying. Check each current offer.

Does it work with Microsoft Teams as well as Slack?

Several do. Runbear, ChatBotKit, Leo, Guru and Agent Barn all list both Slack and Microsoft Teams. Notion's listing mentions Slack. Confirm that the exact channel type you need is supported before you commit.

Can the assistant see private channels and documents?

It depends on how it is configured. Guru says permissions are inherited from the original systems, Leo limits what each employee sees through role-based access, and Runbear uses per-user authorization for connected tools. Review the permissions you grant, and test with a person who should not see certain content.

Can I keep my company data private?

Read each vendor's data and security terms before you connect sources. Agent Barn can be self-hosted, Runbear supports HIPAA only under an eligible Enterprise agreement with a BAA, and Leo lists deployment options including an air-gapped data centre. Involve your security team early.

How long does it take to set up?

Vendors do not publish standard setup times. Cleaning your knowledge sources, setting permissions and testing take longer, and self-hosted options need engineering time.

Compare more options in our Chat assistants category.