3 Ways to Add AI to Freshdesk Without Using Freddy AI

Created time

Aug 17, 2026 09:16 PM

Title length (<60)

Author

Mike Heap

Last optimised

Ecomm?

What do you need before you start in Freshdesk?

⚡

TL;DR: Administrator access to Freshdesk, published solution articles, and about an afternoon. Route A asks nothing of a developer, and that is where most teams land.

Permissions: Administrator is the role you need. Freshdesk's role definitions draw the line at billing. An Administrator configures everything under the Admin tab, and the tier above only adds billing and account management. Check yours at Admin > Team > Roles, and note that where someone holds several roles, the highest privilege wins.

An agent seat for the AI: create a new agent account at Admin > Team > Agents, and use its API key. Replies come from whichever account you connect on, so give us a seat of our own. A shared human seat gets confusing fast.

Plan tier: Freshdesk's pricing page puts Freddy AI Agent on Growth, Pro and Enterprise alike, Freddy AI Copilot on Pro and above as a paid extra, and Freddy AI Insights on Enterprise only, with add-ons managed separately. Which line you land on is usually why you're here.

Data prep: published solution articles. Your existing Automations, canned responses and groups stay untouched.

The Freshdesk New category form for the solutions knowledge base, with a Name field reading Shopping, an empty Description, a portal-visibility chip and Cancel and Save buttons.

What you can skip: the solution-article rewrite, and the audit of every Automation you own. Test on what you have. With no solution articles at all, we draft starter knowledge from your historic tickets instead. The default backfill is the last 5,000.

Time: an afternoon to install and scope, then a week of watching it write private notes.

What are the three ways to add AI to Freshdesk without Freddy AI, and which one should you pick?

⚡

TL;DR: Freshdesk ships AI of its own, and that is the option already sitting in your account. If the piece you need is priced above your plan, or you tried it and it fell flat, there are three other ways in: a Freshworks Marketplace app, a copilot trial your customers never see, and a build of your own against the API.

Freshworks has its own AI for ticketing, and Freshdesk documents it properly. Two reasons people end up here anyway: the piece they wanted sits above the plan they're on, or they tried it and it didn't land.

Our complete guide to Freddy AI covers what it does, and our Freddy AI alternatives roundup ranks who else answers Freshdesk tickets.

Route	Best for	Setup time	Needs a developer?	Trade-off
A. Freshworks Marketplace app	Most teams	About an afternoon	No	You work inside the app's model of routing
B. Copilot / private-note test	Proving it before a customer sees it	About an hour	No	A human still sends it, so handle time drops and the ticket count holds
C. Custom API / webhook build	Non-standard routing, or an in-house model	Days to weeks	Yes	You own it forever, including the `403`s

Route A is the boring-but-effective option: the Freshworks Marketplace. Freshdesk's own description: "The Freshdesk Marketplace has a host of apps that work with Freshdesk to enable you to provide complete customer support." Installing one takes a click and the credentials for the AI's agent seat.

Route B is the same app in private-note mode. It drafts a note your team can send or bin, and no customer sees a word of it. Its case is risk: you prove the thing works on your own tickets before anyone outside the building meets it.

Route C is a ticket-creation Automation with a webhook action. It calls your own service, which replies through the Freshdesk API.

The edges I'd plan for are Freshdesk's rate limits and an API key tied to a single agent seat.

If you can't decide, I'd start on A, leave it writing private notes for a week, and only then widen. Route C is for the rare case where your routing is odd enough that no Marketplace app reaches it.

A five stage rollout sequence for an AI agent in Freshdesk. Install and scope it, run a week of private notes that nobody sends, turn on direct replies for one ticket type, widen a tag at a time, then bring in the next group.

How to add a third-party AI agent to Freshdesk, step by step

⚡

TL;DR: Seven steps: check your role, install the app, sync your solution articles, scope which tickets it sees, spend a week in private notes, set the handoff, then let it reply to one ticket type.

Two questions settle which agent you pick, both answered by the vendor's Marketplace listing before you install anything. Does it install into Freshdesk as a Freshworks Marketplace app, and will it read your solution articles and canned responses as they stand?

Video preview — Connect Your Helpdesk to AI

Running a different helpdesk? We wrote this same walkthrough for Zendesk and Gorgias.

Step 1: Check your role and what the app will ask for

Open Admin > Team > Roles and confirm you're an Administrator. If you can see and edit that screen, you have enough for everything below. Freshdesk resolves multiple roles to the highest privilege among them, so a second, narrower role can't take that away.

Then give the AI its own agent seat at Admin > Team > Agents. Don't lend it yours: we reply from whichever agent account you set the integration up on, so an AI writing under your name is confusing for the customer and impossible to filter on afterwards.

Give that seat Administrator rights too. The API key is per-agent and inherits the role of the agent it belongs to, so an under-privileged seat hands you a key that gets refused later. That 403 is the first thing I'd check when the app won't connect.

Our connection asks for two things: your Freshdesk domain and an API key. The key lives at Profile icon > Profile settings > Your API Key, on whichever agent account the AI will reply from, so fetch it from the seat you just made. Use your original yourdomain.freshdesk.com address here, even if you moved to a custom domain since.

The My Profile Settings page in Freshdesk, with Your API Key and its Reset API Key button outlined in Freshworks' own red callout. The key value itself is blacked out, as are the account name and email.

You're done when the AI's own agent seat exists and you have its API key on the clipboard.

Step 2: Install the app from the Freshworks Marketplace

The path is Admin > Support Operations > Apps, and the last hop depends on your account's build. Freshdesk's own docs currently carry both Get More Apps and Marketplace, so take whichever your screen shows.

Luckily, neither label is wrong, and there's no setting that switches between them, so don't waste ten minutes looking for the one you read about.

Find your AI vendor's app and install it. Freshdesk's own instruction is a single sentence: "Click on Install and enter the necessary credentials and the app will be installed in your helpdesk."

Then configure it: "Under Settings, configure your Freshdesk Domain URL and API Key." Both values come from Step 1. Use the original yourdomain.freshdesk.com address (not the custom one), and the key that belongs to the AI's own agent seat rather than to you.

A Freshworks Marketplace app's Configuration screen on its Settings tab, with Freshworks' own red callouts around the Enter Freshdesk URL and Enter API Key fields above a Verify Your Freshdesk Account button. Both values are blacked out.

Freshdesk's docs add one caveat: "The Apps section in your helpdesk only displays installed apps". So if you go hunting for it before installing it, you'll come up empty.

You're done when the app is listed under your installed apps and its Settings pane has accepted your domain and key.

Step 3: Point it at your knowledge

Three passes, in this order: solution articles, then canned responses, then historic tickets.

Solution articles are the easy half, because you don't write any of them again. Connect your Freshdesk knowledge base as a source and we read what you already have, behind-login articles included, then re-read it every 24 hours. Canned responses are where people go wrong, so budget most of this step for them.

The Freshdesk canned responses admin page, breadcrumb Admin then Canned Response, listing the Personal, General and Escalations folders beside the responses Account Details, Cancel subscriptions and Order status, with Export and New Canned Response controls.

Export them in bulk from Admin > Agent Productivity > Canned Responses. Freshdesk documents the export, and gives its own reason to audit them while you're in there: "minimizing their need to search, check and edit outdated canned responses."

Do not paste that export in raw. Run it through an LLM first and convert it into knowledge. Our canned-response guide carries the prompt, and its first rule does most of the work: convert, never create.

Then historic tickets. We train on them, with a default backfill of the last 5,000.

When an agent answers wrongly, it's almost always the information it was given: a stale solution article, a vague canned response, or two sources that disagree.

Outright invention is rare. A raw canned-response dump is the vague-and-disagreeing failure in one move.

You're done when all three sources show up in the knowledge list.

Step 4: Scope which tickets it sees

You have four levers, all Freshdesk's own:

Groups, at Admin > Team > Groups

Tags, applied from the ticket properties panel and managed at Admin > Agent Productivity > Tags

Views (the queue your team already works from)

A ticket-creation Automation, at Admin > Workflows > Automations > Ticket Creation

Rule order is the part that bites, and Freshdesk says so: "Confirm rule order on Admin > Workflows > Automations > Ticket Creation. With Execute first matching rule, only the first matching rule runs." An older rule sitting above yours will swallow the ticket, and nothing tells you it did.

Your existing Automations stay as they are. Our connection creates its own Automation for New and Updated tickets, and you add conditions to that one to filter what reaches the AI.

Auto-tagging classifies what arrives. Our Guidance feature sets how the AI behaves once a ticket is in scope.

The week-one mistake I'd watch for: teams put in so many escalation, triage and routing rules that the AI never gets a chance at the rest of the queue.

Scope narrow to start. Then widen it a tag at a time.

The Freshdesk Manage Tags admin page, tags sorted by usage with per-tag counts, and a search box typed with refu offering refund, refund failure and Refunds.

You're done when an out-of-scope test ticket lands and nothing touches it.

Step 5: Run it in private-note mode first (Route B)

Freshdesk's word here is private note, and it lives on the ticket action bar: "Private note: Visible only to agents within your organization, as long as they have access." Private threads keep the discussion among agents.

Our connection defaults to exactly this. It writes the reply it would have sent as a note, along with the links it used, so your team can audit the reasoning.

The Reply and Note tab pair at the top of a Freshdesk reply composer, Reply selected and Note sitting beside it, above the From and To rows and the opening line of a draft reply. The sender's address after freshworks and the whole recipient address are blacked out.

If you want a human on every draft, use our AI Copilot Chrome Extension.

In the response composer, click the sparkle icon, then Draft AI agent response. Your agent gets a draft to edit or discard.

Leave it in private-note mode for a full week, and read what it wrote on every ticket. That week is when you fix tone and write the solution articles it turned out to need. When a note looks wrong, ask Echo (our in-dashboard assistant) why the agent answered that way, and which source it used.

You're done when new tickets carry an AI private note, and nothing customer-visible has left Freshdesk.

Step 6: Set the human handoff

Two settings do the work. Can't answer a question hands the ticket to a person the moment the AI has no answer for it. Escalation request escalates only when someone explicitly asks for a person.

I'd run both. It should always be easy to reach a human: they ask outright, the AI can't answer, they're visibly getting annoyed, or the topic is one you've decided belongs with a person. The handover carries a summary too, so your agent picks it up warm.

On the Freshdesk side, an escalation adds the human-handover-requested tag and a note to the ticket. Route those to a human by building an Automation on the Ticket Updates tab that filters on that tag and reassigns the ticket.

A Freshdesk Automation rule builder whose event is Ticket is Updated with Status changed from Any status to Pending, one condition on a company property, and Add watcher as the action.

You're done when a test ticket nobody could answer lands back on a human, carrying the human-handover-requested tag and the summary in a private note.

Step 7: Turn on direct replies for one ticket type

The toggle is Reply directly to customers, in the app's own settings (Freshdesk's own admin screens have no toggle for it). There's a companion setting, Reply to 1st message only, if you'd rather it opened the conversation and then stopped.

Turn off your automatic ticket-creation acknowledgment email before you flip Reply directly to customers. Otherwise the customer gets two emails for one ticket.

Scope it to one tag, in one view, on one group.

By default we answer Email and Web form tickets that are New. On Freshdesk email is usually the first queue you'll point it at.

To roll it back, use the Pause AI agent toggle. Replying to a ticket yourself also stops us on that ticket immediately.

You're done when a ticket on that one tag gets its reply from the AI's own agent seat, and every other view in Freshdesk is still worked by people.

What changes if you build it yourself (Route C)

You build a ticket-creation Automation with a Trigger Webhook action, pointing at your service, which replies through the Freshdesk API.

Auth first: use the Administrator seat you made in Step 1, since the key inherits that agent's role and an under-privileged one returns a 403 (even when the key itself is fine). Fetch the key at Profile icon > Profile settings > Your API Key.

Then rate limits. Freshdesk's API docs are explicit: "Even invalid requests count towards the rate limit." Read X-Ratelimit-Remaining, honor Retry-After, and don't retry blind.

Then the loop: a Ticket Updates rule fires on updates, including the ones your own service makes. Freshdesk's own reopen rule is the clearest example: "Tickets automatically reopen when there's a customer response because of the enabled automation rule - Automatically Reopen tickets after the customer responds (under Admin > Workflows > Automations > Ticket updates tab)." Exclude the AI's agent seat, or its tag, in the rule conditions.

You're done when your endpoint's last call succeeds, the reply lands on the ticket, and nothing re-fires behind it.

How to test it before you let it near customers

⚡

TL;DR: Put it in private-note mode against tickets you have already closed, make it hand one over, prove it can't invent a policy, and record where first response time and CSAT sit before you widen the scope.

Point it at tickets you have already closed, in private-note mode, and read its draft next to the reply your agent really sent. Pass: you can say why each draft differs, and every reason lands on something you can fix inside Freshdesk, a solution article nobody ever wrote, a scope drawn too wide, or a canned response that has gone stale. I'd cover your top few ticket reasons and stop there, because batch size is a judgment call.

Make it hand over. Ask about something your solution articles have never covered. The pass mark is binary. It has to hand the ticket on rather than guess, the summary has to be generated, and the human-handover-requested tag has to land. It shouldn't reply again until you hand it back.

Try to make it invent a policy. Pick an edge case that no solution article of yours addresses. The pass mark is binary again. It either hands the ticket on or says nothing, and no invented policy lands in the thread (that is the one failure you can't take back). Block the categories you can't afford to get wrong at the routing layer in Step 4, so the model never sees them.

Check it respects the scope you set in Step 4. Raise a ticket that should be out of scope, on the wrong group or the wrong tag. Pass is binary: the ticket stays untouched and no private note appears.

Baseline first response time and CSAT before you turn on direct replies. Pass is a dated baseline that covers a full week.

Then comes the go-live gate. I'd hold every one of these four before Freshdesk sends a customer an AI-written reply:

A Freshdesk ticket thread in which an agent has added a private note reading that a refund is to be processed by 9th December, below a forwarded message, with the ticket action bar above showing Reply, Add note, Forward, Close, Merge and Child service task.

Checks 2, 3 and 4 pass on your live instance.

Seven days of private-note drafts are done, and every draft your team binned has a reason you can name.

First response time and CSAT both have a dated baseline on paper.

Direct replies are on for exactly one group or view.

I'd skip the accuracy threshold. One Freshdesk instance differs from the next by enough that a week-one percentage tells you nothing.

What breaks when you add AI to Freshdesk, and how do you tell?

⚡

TL;DR: Most failures trace back to the Automation or the knowledge sync.

Route	Symptom	What's actually happening	Fix
A	The AI never picks the ticket up	A ticket-creation rule runs only when every condition matches, and rules run top to bottom. Tickets converted in from another Freshworks channel can miss the rule entirely	Re-read the conditions against a real ticket, confirm rule order, and check the mailbox's default group at `Admin > Channels > Email`
A	The customer gets two emails the moment you turn on direct replies	Freshdesk's own ticket-creation acknowledgment is still running alongside the AI's reply	Disable the automatic acknowledgment email before enabling direct replies
A	It answers confidently from a stale solution article	The knowledge it read is out of date, ambiguous, incorrect or conflicting	Trace the answer back to its source with Echo, fix or retire the article, and audit canned responses before converting them
C	The API returns `403`	The key is per-agent and inherits that agent's role, so an under-privileged seat is refused even with a valid key	Use the key of an agent with Administrator rights. A `401` is the other problem, a missing or incorrect Authorization header, so check the auth header format before you blame the key
C	Calls start failing under load with `429`	Freshdesk's API docs count every request you send toward the per-minute limit: "Even invalid requests count towards the rate limit."	Read `X-Ratelimit-Remaining`, honor `Retry-After`, and back off rather than retrying blind
C	A ticket-update rule keeps re-firing, on the AI's own update	Ticket Updates rules run on every update, including the ones your service makes	Add a condition excluding the AI's agent seat or its tag, and review the Ticket Updates tab before you ship anything that writes back

The duplicate acknowledgment email and the re-firing update rule are the same failure in different clothes. Something writes back, and a rule you'd forgotten reads that write-back as a customer event.

What should you do next once the AI is live in Freshdesk?

⚡

TL;DR: When the first ticket type holds, add tags and then groups, and fix whatever it couldn't answer at the source.

Widen tags, views and groups one lever at a time. Add one more tag, give it a few days, then bring in the next group. Move two levers in the same week and the number stops telling you which one did it.

We build the review loop for you. The ai-agent-replied and human-handover-requested tags make two ready-made Freshdesk views: one for spot-checking answers, one for everything it handed over.

Work the second list. Whatever it handed over is the queue for your next solution article.

And when the number stops moving, look at your own rules before you look at the model. I'd expect the ceiling to be the routing you built.

If seven steps of wiring sounds like a lot, our Freshdesk integration page covers the same ground with the connection, the tags and the handoff already built. And if the version you put together yourself is already answering tickets, stopping there is a perfectly good outcome.

FAQs

Does Freshdesk have AI?

Yes. Freddy AI for ticketing is Freshworks' own, and Freshdesk documents it in full. Our complete guide to Freddy AI covers what it does and where it stops.

Do I need admin access to do this?

Administrator, yes. Freshdesk's role definitions put billing and account management in the tier above, and none of these steps touches either. If you hold more than one role, the highest privilege applies.

Will this break my existing automations and canned responses?

No. Our connection creates its own Automation for New and Updated tickets, and leaves yours alone. Your canned responses are exported and converted into knowledge, and the originals stay in Freshdesk.

An existing ticket-creation rule sitting above the new one can swallow the ticket first: "With Execute first matching rule, only the first matching rule runs." Check the order before you conclude the AI is broken.

Can I roll it back?

Three ways, and none of them needs a support ticket. Pause the AI agent from the setup settings, take over a single ticket by replying to it yourself, or disconnect us entirely.

Uninstalling a Marketplace app takes the data with it. Freshdesk's Marketplace FAQ is explicit: "Freshworks will delete all data associated with an app after the app is uninstalled."

Rollback	What you do	What it leaves behind
Pause the agent	Flip `Pause AI agent` in your setup settings	The install and its knowledge stay put
Take one ticket back	Reply to that ticket yourself	We stop there, the rest carry on
Disconnect	Unhook us from our dashboard	Freshdesk keeps every ticket and note
Uninstall the app	Remove it from your installed apps	Freshworks deletes that app's data

How long before it starts working on real tickets?

An afternoon to install and scope it. A week of private notes while you read the drafts. Then I'd start with one ticket type on direct replies.

That week is the test plan: private-note drafts on real tickets, a forced handoff, and a scope check.

Does this cost anything on my current plan?

Not on the Freshdesk side for the routes here. Marketplace apps install on your existing plan, and the third-party agent is priced by whoever builds it (us included).

Freddy AI is the one with tiers. Freddy AI Agent runs on Growth, Pro and Enterprise, Freddy AI Copilot needs Pro or above and is charged per agent on top, and Freddy AI Insights is Enterprise only, with add-ons managed separately. Our Freddy AI pricing explainer has the current numbers.

Freddy AI piece	Plans it runs on
Freddy AI Agent	Growth, Pro and Enterprise
Freddy AI Copilot	Pro and above, charged per agent
Freddy AI Insights	Enterprise only

How to Add AI to Freshdesk Without Using Freddy AI

What do you need before you start in Freshdesk?

What are the three ways to add AI to Freshdesk without Freddy AI, and which one should you pick?

How to add a third-party AI agent to Freshdesk, step by step

Step 1: Check your role and what the app will ask for