How to set up Freshservice AI agent: A complete guide for 2026

Table of Contents

• What you'll need before starting
• Step 1: Enable Freddy AI Agent in your admin settings
• Step 2: Configure your knowledge sources
• Step 3: Set up your AI Agent persona and responses
• Step 4: Deploy to Slack or Microsoft Teams
• Step 5: Test and refine your AI Agent
• Common setup issues and how to avoid them
• Getting the most from your AI Agent

Setting up an AI agent for your IT service desk sounds complicated, but it doesn't have to be. Freshservice's Freddy AI Agent can handle employee questions, route requests, and even resolve issues without human intervention. The catch? You need to configure it properly first.

This guide walks through the complete Freshservice AI agent setup process, from prerequisites to deployment. Whether you're rolling it out on Slack, Microsoft Teams, or your support portal, you'll find step-by-step instructions that actually work.

What you'll need before starting

Before diving into configuration, make sure you have the basics covered. Freddy AI Agent isn't available on every plan, and you'll need proper permissions to get started.

Freshservice Enterprise plan. Freddy AI Agent is only available on the Enterprise tier. Starter ($19/agent), Growth ($49/agent), and Pro ($99/agent) plans don't include AI agent capabilities. If you're on a lower tier, you'll need to upgrade first.

Admin permissions. You'll need admin access to your Freshservice instance. For Microsoft Teams integration specifically, you also need Global Admin rights in Teams.

Knowledge base content. Freddy AI Agent learns from your existing content. Have your solution articles, FAQs, and documentation ready. The AI can also pull from SharePoint and uploaded files (.txt, .docx, .pdf up to 35MB each).

Session allocation. Each Enterprise license includes 1,200 Freddy AI Agent sessions per year. A session is any interaction a unique user has with the AI within a 24-hour period. Plan your rollout accordingly.

Channel requirements. If you're deploying to Slack, you'll need workspace admin access. For Teams, it's a 1-to-1 mapping with your Freshservice instance (multi-workspace isn't supported for Teams).

Step 1: Enable Freddy AI Agent in your admin settings

Let's get Freddy AI Agent turned on. This happens in your Freshservice admin panel.

Navigate to Admin → Global Settings and search for "Freddy." Select it from the results. You'll see toggle switches for each channel where you can deploy the AI agent.

Turn on the channels you want to use:

• Slack (requires ServiceBot installation first)
• Microsoft Teams (requires ServiceBot installation first)
• Support Portal (built into your Freshservice portal)
• Email Bot (auto-responds to email queries)

Don't worry about perfect configuration yet. You can adjust settings after enabling the basic functionality. The key is getting the foundation in place so you can start testing.

Step 2: Configure your knowledge sources

Freddy AI Agent is only as good as what it knows. This step connects the AI to your institutional knowledge.

Connect Freshservice solution articles. These are automatically available if they're published in your knowledge base. The AI processes the first 50 inline images per article and can reference them when answering questions.

Add SharePoint content. If your organization uses SharePoint for documentation, connect those libraries. Freddy can search across both Freshservice and SharePoint to find answers.

Upload files. You can add .txt, .docx, and .pdf files (max 35MB each, up to 200 files per bot). This is useful for policies, procedures, or training materials that aren't in your knowledge base yet.

Add public URLs. Freddy can learn from publicly available web pages (up to 10 URLs per agent, max 3,000 pages total). Only static text content is processed, not videos or animations.

Set up custom Q&A pairs. For common questions that don't fit neatly into articles, create direct question-and-answer pairs. This helps with specific terminology or processes unique to your organization.

Be patient after adding sources. Processing takes anywhere from 1 hour to 24 hours depending on content volume. You'll get an email when URL learning is complete.

Step 3: Set up your AI Agent persona and responses

Now let's make Freddy sound like part of your team, not a generic chatbot.

Define the persona. Give your AI agent a name and avatar that match your brand. Set your industry, offerings, and key terminology so Freddy understands your business context.

Configure tone and behavior. Write plain-English instructions guiding how Freddy should interact. For example: "Be professional but friendly" or "Always offer to escalate billing issues to a human."

Customize greeting messages. Set the introductory message users see when they start a conversation. You can use placeholders for customer name, email, and other fields.

Set up escalation messages. Define what Freddy says when transferring to a human agent. Include context about what was already discussed so the agent doesn't start from scratch.

Configure failure messages. When Freddy can't answer a question, the failure message should set expectations and offer alternatives (like creating a ticket or speaking to an agent).

Enable spam and out-of-scope handling. This automatically deflects irrelevant or unclear queries without triggering fallbacks. It reduces noise for your human agents.

Step 4: Deploy to Slack or Microsoft Teams

This is where Freddy AI Agent becomes available to your employees in the tools they already use.

For Slack:

1. Go to Admin → Channels → ServiceBot for Slack
2. Click Install and authorize permissions with your Slack workspace credentials
3. Select which events should trigger notifications through ServiceBot
4. Click Save and Continue
5. Create a test group to validate Freddy before full rollout
6. Add agents and requesters to the test group
7. Name the channel where Freddy will be enabled for testing
8. Click Launch now to enable for select users
9. Once satisfied, click Launch to everyone

For Microsoft Teams:

1. Go to Admin → Channels → ServiceBot for Microsoft Teams
2. Click Install and authorize with your Microsoft admin account
3. Select notification events and save
4. Create a test group (similar to Slack process)
5. Launch to test group first, then everyone

Important notes:

• Slack Enterprise Grid supports multiple workspaces; Teams does not
• Forms must be simple to render in chat (less than 8 fields, fewer than 100 dropdown choices total, no dynamic sections)
• For Teams, you must be logged into the same account you're integrating with

Step 5: Test and refine your AI Agent

Don't skip testing. A poorly configured AI agent frustrates users more than no AI at all.

Use the preview environment. Before going live, test conversations in the preview mode. Try common questions, edge cases, and ambiguous queries to see how Freddy responds.

Review conversation logs. After launch, regularly check the Analyze tab. Look at:

• Total conversations and resolution rates
• Agent transfer rates
• Unanswered queries
• Unhelpful responses (marked by users)

Identify knowledge gaps. When Freddy can't answer a question, that's a signal. Either add the content to your knowledge base or create a custom Q&A pair.

Iterate on responses. If users consistently mark responses as unhelpful, review and refine your knowledge base content. The AI learns from corrections, so agent edits improve future responses.

Monitor session consumption. Keep an eye on your 1,200 sessions per agent per year. If you're burning through them quickly, you may need to purchase additional packs (100 sessions per pack) or optimize which queries go to the AI.

Common setup issues and how to avoid them

Even with clear instructions, things go wrong. Here are the most common pitfalls and how to avoid them.

Form compatibility errors. Complex forms don't work in Slack or Teams. If users get redirected to the portal unexpectedly, check your forms against these limits:

• Less than 8 fields
• Fewer than 100 dropdown choices across all fields
• No dynamic sections, dependent fields, or multi-select fields
• No mandatory attachments for service requests

Permission problems. Teams setup requires Global Admin rights, not just Teams admin. If installation fails, verify your Microsoft permissions.

Knowledge source delays. Don't expect immediate results after uploading content. Processing takes 1-24 hours. Plan your rollout timeline accordingly.

Session confusion. Remember that a "session" is per user per 24-hour period, not per interaction. One user chatting with Freddy all day counts as one session, not dozens.

Multi-workspace limitations. If you have multiple Slack workspaces, you need Enterprise Grid for multi-workspace support. Teams only supports 1-to-1 mapping.

Getting the most from your AI Agent

Once Freddy AI Agent is running, optimize for better results.

Prepare your knowledge base. The AI is only as good as your content. Organize articles with clear titles, use consistent terminology, and keep information current. Break long articles into focused pieces covering single topics.

Set up workflows. Extend Freddy's capabilities with automated workflows. Common use cases include password resets, software provisioning, and account unlocks. These reduce the load on your human agents.

Train your team on handoffs. When Freddy escalates to a human, the full conversation history transfers too. Make sure agents know to review the context before responding. This prevents users from repeating themselves.

Monitor performance metrics. Track resolution rates, user satisfaction, and session consumption. Use these insights to justify expansion or identify areas needing improvement.

Plan for scale. Start with a pilot group, then expand department by department. This controlled rollout lets you catch issues early and build internal advocates.

If you're exploring alternatives or need AI capabilities beyond what Freddy offers, eesel AI integrates with Freshservice to provide additional automation options. Our approach treats AI as a teammate that learns your business and escalates intelligently based on rules you define in plain English.