Fini (usefini.com) runs inside the helpdesk you already use: connect Zendesk, Intercom, Front, Gorgias, HubSpot, Microsoft 365, Salesforce, LiveChat, or Slack, and the AI agent reads incoming tickets and replies in that tool, with no migration. Your tickets, inboxes, and team stay where they are; Fini becomes the first layer that answers what it can and hands the rest to your people with the conversation and reasoning trace attached. This page explains the pattern once, across all nine helpdesk integrations: what Fini reads and writes in your helpdesk, how it decides between a reply, an internal note, and silence, how handoff to humans is marked, how brands and inboxes route to agents, and how the chat widget escalates into a helpdesk. Each integration’s own setup page has the exact connect flow.

​
How it works

  1. A customer writes in through a channel your helpdesk already owns: an email ticket, a chat conversation, a Front inbox, a shared mailbox, or a Slack channel.
  2. The helpdesk notifies Fini. Depending on the integration this is a webhook and trigger Fini creates during authorization (Zendesk, Gorgias), the integration’s own event delivery, or, for Salesforce, two record-triggered Flows you build.
  3. Fini picks the agent mapped to that brand, inbox, channel account, mailbox, or channel, and runs it with the same knowledge, prompts, and Rulebook it uses on every other surface.
  4. Reply Rules decide whether the agent posts a direct reply, leaves an internal note for your team, or stays silent.
  5. If the conversation needs a person, Fini marks it as escalated in the helpdesk (with a tag, ticket field, or ticket property, depending on the tool) and your team picks it up in the views and queues they already use.
The same flow, message by message, including the three Reply Rules outcomes and an escalation: The configuration lives on the Fini side once the helpdesk is authorized. You don’t rebuild queues, move historical tickets, or change the tool your team replies from.

​
What stays where

​
The first-layer pattern

Most production deployments put Fini in front of every customer interaction. The agent triages, answers the high-volume long tail (status checks, policy questions, how-tos, account workflows you’ve built in Rulebook), and escalates the rest. Because the agent uses one set of knowledge and rules regardless of channel, a cancellation-policy question gets the same answer by Zendesk email, Intercom chat, or the widget on your site. See The first-layer pattern for the full rationale. Fini’s stated rollout timeline for this pattern is a knowledge agent live on day 1, agentic workflows by day 14, and full autonomy by day 30.

​
What Fini reads and writes in your helpdesk

Each setup page lists the exact permissions and side effects. Across integrations they fall into the same groups: Zendesk, Intercom, Front, Gorgias, and HubSpot each publish Fini’s full permission set on their app marketplace listing, linked from the setup page. Review it before authorizing.

​
Reply, internal note, or silence

Every integration hands the final posting decision to Reply Rules (Rulebook → Reply Rules). The three cards are Direct Reply, Internal Comment, and No Reply, and the stricter one wins when more than one matches: No Reply beats Internal Comment, which beats Direct Reply. Reply Rules match on system fields that come from the helpdesk, plus your tags and user attributes:
  • Integration Provider and Channel scope a rule to, for example, Zendesk chat only.
  • Human Agent Assigned is True when a teammate owns the conversation. A No Reply rule on Human Agent Assigned Equals True is the standard way to stop the agent talking over a person.
  • Escalated Conversation is True once the conversation has been escalated.
Rules are set once and apply across every connected integration, so a “stay silent when a human is on it” rule covers Zendesk, Intercom, Front, and the rest at the same time. Zendesk, Intercom, Front, Gorgias, HubSpot, Salesforce, and LiveChat also have Reply Settings with a response delay in seconds per channel (Slack has no Reply Settings section). 0 replies as soon as the agent has an answer; a higher value gives a human a chance to take over first.

​
How handoff to humans works

When the agent can’t or shouldn’t resolve a conversation (a policy says so, confidence is low, or the customer asked for a person), it escalates. In a helpdesk integration, escalation means two things:
  1. The conversation is marked as escalated in your helpdesk so your existing views, triggers, and assignment rules can pick it up. The marker differs by tool: a custom ticket field in Zendesk, the fini_escalated_human_agent tag in Intercom, Front, Gorgias, and LiveChat, a custom ticket property in HubSpot, and the Fini_Transfer__c checkbox on the Case in Salesforce.
  2. The agent stops replying while a human handles it, which you enforce with Reply Rules on Human Agent Assigned and Escalated Conversation.
Your team keeps working in the helpdesk. The full Fini trace for the conversation (planning, retrieved knowledge, executed rules and attributes) stays available in Inbox, and the Source URL in Conversation metadata links straight to the live ticket. To keep Fini out of one conversation without disabling the integration, set the integration’s skip marker, for example the Fini Skip Webhook tag in Front, Gorgias, or Intercom, or the Fini_Skip_Webhook__c checkbox on a Salesforce Case.
For escalation triggers inside the agent (which topics always go to a human), see Controlling when the agent escalates in Prompts. In workspaces with native ticketing, widget and native email escalations can instead become native tickets in Inbox, routed to your team by Agent groups.

​
Multiple brands, inboxes, and workspaces

You can run one agent across everything or map different agents to different parts of your helpdesk. The routing unit depends on what the helpdesk exposes:
  • Zendesk routes by brand, separately for email and chat. Leaving Select brands empty makes an agent answer every brand on that channel.
  • Intercom routes by brand, but Intercom has no built-in brand field: Fini reads a custom conversation attribute named Brand. Always configure a wildcard agent (no brands selected) to catch conversations without a matching value.
  • Front routes by inbox. An inbox maps to one agent per channel; an agent with no inboxes selected is the wildcard.
  • HubSpot routes by channel account, separately for email and chat. There’s no per-brand split.
  • Microsoft 365 routes by shared mailbox: up to 50 shared mailboxes per connection, each with its own agent and up to 20 monitored folders.
  • Slack routes by channel: one agent can cover many channels, each channel maps to at most one agent.
  • Gorgias, Salesforce, and LiveChat use one agent per channel with no brand or inbox split.
To confirm (internal, remove before publish): The Channel overview card says Intercom works “across one or many workspaces”, but the Intercom setup page describes selecting one workspace per authorization. Can one Fini workspace connect several Intercom workspaces at once, and if so, how is each routed?
Separately, one Fini workspace can run several agents, each with its own knowledge, configuration, and deployments, so per-brand routing can point at genuinely different agents rather than one agent with branching prompts.

​
Widget escalation into your helpdesk

The most common combined setup: the chat widget answers on your site or app, and when it can’t help, the conversation hands off as a real ticket in your helpdesk. The customer sees the handoff inside the widget; your team picks it up in their usual tool. Five helpdesks can be widget escalation destinations today: Zendesk, Front, Gorgias, Salesforce, and HubSpot. Intercom, LiveChat, and Slack work as direct integrations only.
1

Authorize the destination helpdesk

Connect it on its Deploy page. If you only use it as a widget destination, you can skip that page’s agent routing and reply settings. For Salesforce, widget-only use doesn’t need the manual Flow setup.
2

Create a Business Rule

In Rulebook → Business Rules, click Use Default Rule to start from a destination template, or Create Custom for a behavior tree. Business Rules run for source Widget on trigger On Escalation.
3

Map fields, assign bots, and test

Map required fields such as Interaction History Transcript, Interaction ID, and Formatted User Attributes, assign the rule to the widget’s bot, and run Test before relying on it.
What each destination creates is covered on the Widget page: a Zendesk ticket (with widget file uploads forwarded as native attachments on Chat escalations), a Front conversation on an email channel, a Gorgias ticket, a Salesforce Case that follows your assignment rules, or a HubSpot ticket, native Help Desk thread, or support-form submission. If no Business Rule applies, no destination ticket is created. See Business Rules.

​
Comparison across helpdesks

Built from each integration’s setup page. “Not documented” means the setup page doesn’t describe it; ask your Fini contact before relying on it. Search the same comparison, or filter to the helpdesks that can receive widget escalations. The table above links to each setup page.
To confirm (internal, remove before publish): The Microsoft 365 and Slack pages don’t describe how an escalation is marked, whether internal notes are supported (Microsoft 365), or whether reply delays exist (Microsoft 365). What happens in a shared mailbox, and in a Slack channel, when the agent escalates to a human?

​
Setup effort and controls

​
Integrations without a setup page

The Deploy page also shows cards for Freshdesk (“Personalized AI responses and effortless handoff for Freshdesk”), Deskpro (“Personalized AI responses and effortless handoff for Deskpro tickets”), and Discord (“AI-driven customer support within your Discord community”). These don’t have setup pages in the docs yet; talk to your Fini contact before planning a rollout on them.

​
Rolling out on a helpdesk

A low-risk order that uses only the controls above:
1

Create and train the agent

Build the agent in Agent home and connect its knowledge, including past tickets from the helpdesk you’re connecting.
2

Authorize the helpdesk and map one slice

Map the agent to one brand, inbox, or channel first, and leave the rest unmapped or covered by a No Reply rule.
3

Start with internal notes

An Internal Comment rule lets the agent draft notes your team reads before anything reaches a customer. Review them in the helpdesk and in Inbox.
4

Add the safety rules

Add a No Reply rule on Human Agent Assigned Equals True so the agent never talks over a teammate, and Internal Comment rules for sensitive topics.
5

Switch to direct replies and widen

Turn on Direct Reply for the slice, lock fixes in with the Test Suite, then extend routing to more brands, inboxes, or channels. Track results in Analytics.

​
Common questions

No. Fini connects to the helpdesk you use and works on its tickets and conversations in place. Your team keeps replying from the same tool.
No. Helpdesk conversations are read-only mirrors in Inbox, used for tracing, replay, feedback, and test sets. Replies happen in the helpdesk. Only widget, Inbox-UI, and native email conversations can be answered from Inbox.
Not if you add the standard No Reply rule on Human Agent Assigned Equals True (and, if you want, Escalated Conversation Equals True). Response delays in Reply Settings also give teammates a window to take over first.
Yes. Surfaces aren’t mutually exclusive, and the same agent (or different agents) can answer on several integrations at once. See Composing surfaces.

Channel overview

Every surface Fini deploys to, and how they compose.

Reply Rules

Decide when the agent replies, leaves an internal note, or stays silent.

Business Rules

Build the widget escalation into your helpdesk.

Inbox

Trace every helpdesk conversation, step by step.