AI Inbox

What is the AI Inbox?

The AI Inbox is a human escalation workspace built directly into ChatSpark. When your AI agent determines it needs human help, it opens an Assist and hands the conversation to your team.

Before your team member reads a single message, the AI has already prepared:

• A plain-language summary of the conversation
• Extracted customer details such as name, email, and order number
• A priority level based on urgency signals in the conversation
• A category label such as billing, refund, or technical
• A suggested reply your agent can use or edit
• An explanation of why the AI escalated

This is different from a traditional ticketing system. Your team does not manage a queue of raw transcripts. They see the decision they need to make, with everything already organized.

How It Works

1. Trigger detection: A customer says something like “talk to a person” or “live agent” and your AI agent detects the escalation intent.
2. Information collection: The AI gathers the customer's name and email through a short form or natural conversation flow, depending on the channel.
3. Confirmation: The customer confirms their details before the Assist is created.
4. Assist creation: The AI analyzes the full conversation and generates a summary, priority, category, suggested reply, and tags. An Assist is created in your AI Inbox.
5. Team notification: Your team receives an email with the AI summary and a direct link to the Assist. With Enterprise AI Routing, authorized members of the selected department receive the new Assist notification. After a team member claims it by responding, customer reply notifications go to that owner, subject to notification preferences.
6. Customer confirmation: The customer receives a confirmation email with a reference number and a note that your team will follow up.

Setting Up the AI Inbox

1. Go to AI Actions in the left sidebar.
2. Find ChatSpark AI Inbox in the Support category.
3. Click Enable on the agent you want to connect.
4. Enter your support team name and the email address you want outbound emails sent from. This is the address your customers will see in their inbox.
5. Choose your email provider from the dropdown. Supported providers are Brevo, SendGrid, Postmark, and Resend. Paste in the API key from your chosen provider. This keeps email delivery tied to your account and your domain reputation.
6. Add your support team notification email. This is where new Assist notifications are sent. Separate multiple addresses with commas.
7. Optionally add your company address in the email footer field.
8. Save the action. The AI Inbox menu item will appear in your sidebar.

The AI Inbox View

The AI Inbox has three panels working together.

Assist list

The left panel shows all your Assists organized by status: Open, Pending, and Closed. Each row shows the customer name, a one-line AI summary preview, a priority badge, and the time the Assist was created. A green dot appears when there is an unread customer reply.

Conversation and reply

The center panel shows the full conversation transcript from the chat session, followed by any internal notes and customer email replies. At the bottom you can reply directly to the customer via email, or leave an internal note that is only visible to your team.

AI Intelligence panel

The right panel is where the AI does the work. It contains:

• Summary: A 2 to 3 sentence plain-language summary of the conversation
• Customer Details: Name, email, and any identifiers the AI extracted from the conversation
• Suggested Reply: A ready-to-send AI draft that appears pre-filled in the reply box the moment you open an Assist
• Assist Info: Priority, category, and AI-generated tags
• Previous Assists: Any prior Assists from the same customer email

AI Reply Refinement

When you open an Assist, the AI-generated reply is already waiting in the reply box, highlighted with an AI draft indicator. Your agent can send it as written, edit it directly, or refine it with one click using the options directly below the reply field.

• Regenerate: Get a fresh suggestion with a different approach
• Shorter: Condense the reply to the essentials
• More formal: Adjust the tone for professional communication
• More casual: Adjust the tone for a friendly, conversational reply

The AI draft indicator clears as soon as your agent starts typing, so there is no confusion about which parts are AI-generated and which have been edited. Switching to the Internal Note tab keeps a clean, separate field for team-only messages.

Email Reply Threading

Every email sent from the AI Inbox includes a reply address managed by ChatSpark. When a customer replies to any email in the thread, their reply is added to the Assist automatically and your team is notified.

The full thread appears in Gmail and Outlook as a native email conversation. Customers do not need to log into a portal or open a new chat. They just reply to the email.

Conversation Continuity

When an Assist is closed, the resolution is stored and linked to the customer's chat history. The next time the same customer starts a conversation with your AI agent, the AI knows what was resolved and can reference it naturally.

Training Your AI from Resolved Assists

Every closed Assist is a training opportunity. Close one and ChatSpark prompts you to add the resolution to your AI with one click. Each question the customer asked is paired with the corresponding reply and saved as a separate training entry. Personal details are stripped before anything is saved. Everything indexes immediately, so the next time a customer asks something similar your AI handles it directly.

Customer Satisfaction

When closing an Assist, you can optionally send the customer a one-click satisfaction survey. Results appear in your reports and negative responses are flagged in the AI Intelligence panel as training signals.

Team Access

Any team member with the Admin role or above can access the AI Inbox by default. If you want to give a team member AI Inbox access without access to agent settings, analytics, or billing, use the Support role.

Support members log in directly to the AI Inbox. They can view assists, reply to customers, add internal notes, and update status and priority. They cannot access agent configuration, training data, or any other account area.

To invite a Support member, go to Team and Security and send an invitation with the Support role selected. See the Team and Security documentation for step-by-step instructions.

Support and Manager role users have an Available or Offline status shown in the sidebar. On Enterprise plans, an Owner or Admin can control whether a Support member can set their own availability status or whether it is managed from the Management view.

AI Routing for Enterprise

AI Routing reduces the work of sorting requests, finding the right person, and coordinating follow-up. When your AI Agent needs human help, it uses the unresolved request and conversation context to select a department and a suitable team member. The summary and suggested reply give that person a starting point for their response.

Departments are shared across your account. Each AI Agent has its own allowed departments, fallback department, and routing mode. For example, a returns AI Agent can route to Returns and Billing while a technical AI Agent routes to Product Support.

Set up departments and member profiles

1. Open AI Inbox, then Settings, then AI Routing. Owners and Admins can also open the AI Routing tab in Team & Security.
2. Create each department. Describe what it handles, add example inquiries, and choose its team members. A person can belong to more than one department.
3. Complete member profiles with position, responsibilities, expertise, languages, and capacity. An authorized administrator can also record required qualifications and approval or escalation authority.
4. Allow AI Routing for the account. Select an AI Agent, choose its allowed departments, and save a Fallback and triage department. This department receives unclear requests and work awaiting routing review.
5. Choose Suggest or Automatic, then click Save AI Agent routing.
6. Use Test an inquiry with a sample customer message or voicemail transcript. Review the department, suggested member, and initial notification audience before using the settings with new work.

Choose a routing mode

• Off: New Assists use the ordinary AI Inbox workflow.
• Suggest: New work stays in the fallback and triage department while AI proposes a route. An authorized reviewer can accept it, choose another department, or keep the triage route.
• Automatic: AI applies the route within the AI Agent's allowed departments. If the context is insufficient or the decision cannot be completed, the saved fallback receives the work.

Explicit routing rules take precedence over contextual selection. AI never adds a department outside the AI Agent's allowed set. If no qualified, available member has capacity, the Assist remains visible in the department for follow-up rather than bypassing its requirements.

Member matching, round robin, and ownership

Within the selected department, AI considers relevant expertise, responsibilities, language, availability, and workload. Required skills and authority determine who can handle the request. Round robin distributes initial responsibility among equally suitable members.

The selected member is initially responsible, but the Assist is not yet claimed. Authorized department members receive the new Assist notification, subject to their notification preferences. The first eligible person to successfully send a customer reply claims it, even if another member was selected initially. Opening the Assist, reading the draft, or adding an internal note does not claim it.

Later customer replies notify the claimed owner. AI prepares an updated draft using the latest context. If the conversation changes while someone is editing, they review the new context before sending. Their edits remain available. Claimed work keeps its owner until an authorized transfer or a configured escalation changes responsibility.

Use the department filter to focus the Assist list. The Assist's routing details show its department and whether responsibility is provisional or claimed. When correcting a route, record the reason so the team can review repeated corrections later.

Response targets and escalation

In the AI Agent's routing settings, enable Set customer response targets. Set separate first and subsequent response targets, a timezone, business hours, and holidays. Add escalation stages with the appropriate delay, department notification, or transfer action.

When a response is overdue, the saved policy can notify the configured department or transfer the Assist to an eligible member. A new customer message does not restart an already unanswered deadline. Moving unanswered work to Pending does not stop that deadline, and closing it without a response is recorded separately from meeting the response target.

Customer-tier policies can raise priority and shorten the first response target. An authorized team member must verify the tier against a customer record reference. A name, email address, or caller number alone does not establish a tier. Expired verification no longer adds priority weight.

Review routing improvements

Open Reviewed routing improvements and select Review recent corrections. Repeated corrections can produce proposed changes to department responsibilities or example inquiries. Review the evidence, edit the proposal, and test it before approving. You can reject a proposal or restore the previous configuration when available. Suggestions change future routing only after approval.

You can also ask AI Operator to review or update departments and routing settings. Operator follows the same permissions and confirmation steps as the dashboard.

Check routing health

Routing health shows configuration issues, delayed work, failed notifications, inbound emails awaiting review, and replies whose delivery outcome needs checking. If delivery is uncertain, confirm the provider result before retrying so the customer does not receive a duplicate response. Recovery actions require the appropriate permission and a recorded reason.

Management

Management is an Enterprise-only view giving Owners, Admins, and Managers a live picture of team workload, queue health, and agent availability. Access it from the Management item in the AI Inbox sidebar after Reports.

Team Workload

Each team member with AI Inbox access has a workload card showing their open Assist count, average resolution time, CSAT score, and current availability. If an agent has significantly more open Assists than the team average, their card shows a high load indicator. Toggle any agent Available or Offline directly from their card.

Availability and Schedule

Expand any agent card to set a weekly work schedule. When the schedule editor is open, a sparkline shows when Assists typically arrive over the last 30 days to help you set coverage times that match actual demand. Agents can be toggled Available or Offline manually at any time regardless of their schedule.

Queue Health

The queue section shows the number of unassigned open Assists, the age of the oldest open Assist, and a 7-day backlog trend.

Reassign Assist

A list of unassigned Assists lets authorized team members assign work to eligible members. With AI Routing, reassignment also respects department membership, required qualifications, availability, and capacity.

AI Inbox Integrations

AI Inbox connects to your existing tools to create Assists automatically. Telephony integrations pull voicemail transcripts as they arrive. Migration importers bring open tickets from a help desk platform when your team switches to AI Inbox.

All integrations are configured per AI Agent in AI Inbox then Settings then Integrations.

Telephony and Voicemail

Telephony integrations create an Assist when a caller leaves a voicemail. Available transcripts provide context for the summary and callback preparation. On Enterprise, these Assists follow the AI Agent's routing settings. If a transcript is delayed or unavailable, the fallback department receives the request. Later transcript context can support a routing suggestion without silently moving a claimed conversation.

Each connected line gets a Brand Label. That label is applied to every Assist from that number so your team can filter by brand, department, or line at a glance.

Dialpad

Connect one or more Dialpad phone numbers to your AI Inbox. When a caller leaves a voicemail, the transcript is pulled in automatically and a new Assist is created for your team.

You can connect multiple numbers from the same Dialpad account. Each number has its own brand label and target scope.

• Brand Label: The label applied to every Assist from this number. Creates or matches an existing Inbox Label in your label settings.
• Target Type: The Dialpad target to monitor. Options are User, Call Center, and Department.
• Target ID: The numeric ID for the target found in your Dialpad admin portal.

1. Go to AI Inbox then Settings then Integrations.
2. Click Configure on the Dialpad card.
3. Enter your Dialpad API key, phone number, brand label, target type, and target ID.
4. Click Connect Line. The line becomes active immediately.

RingCentral

Connect one or more RingCentral extensions to your AI Inbox. When a caller leaves a voicemail, the transcript is pulled in automatically and a new Assist is created for your team.

Each extension you connect gets a Brand Label. That label is applied to every Assist created from that extension so your team can filter by brand, department, or line at a glance.

• Client ID: Your RingCentral app Client ID from the Developer Console.
• Client Secret: Paired with the Client ID. Keep it private.
• User JWT: A user-level JWT token from the Developer Console under App Credentials.
• Extension ID: The RingCentral extension to monitor. Use ~ for the current user or enter a specific extension ID for a call center or department.
• Brand Label: The label applied to every Assist from this extension. Creates or matches an existing Inbox Label.

1. Go to AI Inbox then Settings then Integrations.
2. Click Configure on the RingCentral card.
3. Enter your Client ID, Client Secret, User JWT, extension ID, brand label, and phone number.
4. Click Connect Line. The line becomes active immediately.

Twilio

Connect a Twilio account to your AI Inbox. When a caller leaves a voicemail on your Twilio phone number, the recording and transcript are pulled in automatically and a new Assist is created for your team.

• Account SID: Your Twilio Account SID from the Twilio Console dashboard.
• Auth Token: Your Twilio Auth Token from the Twilio Console dashboard.
• Phone Number: The Twilio phone number this line monitors. Used for identification only.
• Brand Label: The label applied to every Assist from this number. Creates or matches an existing Inbox Label.

1. Go to AI Inbox then Settings then Integrations.
2. Click Configure on the Twilio card.
3. Enter your Account SID, Auth Token, phone number, and brand label, then click Connect Line.
4. Copy the Webhook URL displayed on the connected line card.
5. In your Twilio voicemail flow, send recording and transcription callbacks to the copied Webhook URL.

Aircall

Connect one or more Aircall numbers to your AI Inbox. When a call ends as a voicemail, the transcript is fetched automatically and a new Assist is created for your team.

• API ID: The API ID generated in Aircall under Dashboard then Integrations then API Keys.
• API Token: Paired with the API ID. Keep it private.
• Phone Number: The Aircall number this line monitors. Used for identification only.
• Brand Label: The label applied to every Assist from this number. Creates or matches an existing Inbox Label.

1. Go to AI Inbox then Settings then Integrations.
2. Click Configure on the Aircall card.
3. Enter your API ID, API Token, phone number, and brand label.
4. Click Connect Line. The webhook is registered automatically.

Quo

Connect your Quo workspace to AI Inbox. When a caller leaves a voicemail, the transcript is pulled in automatically and a new Assist is created. One webhook covers your entire workspace, with optional per-number filtering.

• API Key: Your Quo API key from Settings then API.
• Phone Number: Optionally filter to a specific number in your workspace. Leave blank to receive voicemails from all numbers.
• Brand Label: The label applied to every Assist from this workspace. Creates or matches an existing Inbox Label.

1. Go to AI Inbox then Settings then Integrations.
2. Click Configure on the Quo card.
3. Enter your API Key and brand label.
4. Optionally select a phone number from the dropdown to filter by a specific number.
5. Click Connect Line. The webhook is registered automatically.

Vonage

Connect a Vonage Application to your AI Inbox. When a caller leaves a voicemail recording, a new Assist is created automatically. Vonage webhook URLs are configured in the Vonage Dashboard rather than registered programmatically.

• Application ID: Your Vonage Application ID from the Vonage Dashboard under Applications.
• Private Key: The private key downloaded when you created the Vonage Application. Paste the full PEM contents.
• API Key and Signature Secret: Required to verify incoming events. Use the signature secret from your Vonage account.
• Caller ID: The authorized phone number used for outbound callbacks.
• Brand Label: The label applied to every Assist from this application. Creates or matches an existing Inbox Label.

1. Create a Vonage Application in the Vonage Dashboard with Voice capability enabled and download the private key.
2. Go to AI Inbox then Settings then Integrations.
3. Click Configure on the Vonage card.
4. Enter your Application ID, private key, API key, signature secret, caller ID, and brand label.
5. Click Connect Application, then copy the Webhook URL shown on the card.
6. In the Vonage Dashboard, open your Application and set the copied URL as the Event URL.

Connected Dialer and Callback Preparation

Follow up on a voicemail from the Assist using your connected phone provider. Click the phone icon in the Assist header to open the dialer panel. The customer phone number is pre-filled when available. If you have more than one connected line, select the line to use for the callback.

AI prepares a callback brief from the available conversation, voicemail, and knowledge sources. Record the actual call outcome and notes so the next person has the follow-up context. Opening the dialer does not claim an Assist. With AI Routing, a confirmed connected callback can establish ownership. An unanswered attempt does not count as responding to the customer.

Supported Providers

RingCentral and Aircall open their workspaces inside the dialer panel. Twilio and Vonage use an in-browser dial pad with microphone access. Dialpad and Quo open their connected applications. Call controls, sign-in requirements, and transcript availability depend on the provider and your account configuration.

App Setup

Open the integration card in AI Inbox then Settings then Integrations and expand the Dialer section.

• RingCentral: Enter a Client ID and Client Secret from a RingCentral app created in the Developer Console.
• Twilio: Enter your TwiML App SID, API Key SID, and API Key Secret. Use an authorized caller number and set the TwiML App Voice URL to the answer URL shown in setup using POST.
• Aircall: Enable the dialer and sign in to your Aircall workspace when prompted.
• Vonage: Configure the Voice application, private key, signed event credentials, and authorized caller ID, then enable the dialer.

Per-User Setup

RingCentral requires each team member to connect their own account. Once the app is configured, click Enable Dialer in the Assist header and authorize your RingCentral account in the popup. Other providers may require workspace sign-in or microphone permission. Existing AI Inbox access and routing requirements still apply when starting a callback.

Callback outcomes and follow-up

Save the outcome and notes after the callback. AI uses the recorded context to prepare follow-up guidance and a suggested email when the customer has an email address. Review it before sending. Transcripts and automatic call details are included when the connected provider supplies them.

Help Desk Migration

Migration importers let you bring open and pending tickets from your current help desk into AI Inbox as Assists when your team switches platforms. Each importer includes full conversation history so your team picks up exactly where things left off.

Imports run on demand from the integration drawer. Tickets already imported are skipped automatically, so you can run the import multiple times safely.

HappyFox

Import open and pending tickets from HappyFox into AI Inbox as Assists, including full conversation history. Use this when migrating from HappyFox so your team can continue working from AI Inbox without losing any context.

Each import run brings in up to 200 tickets. You can run it again to import the next batch. Tickets already imported are skipped automatically, so running it twice is safe.

• Subdomain: The first part of your HappyFox URL. For acme.happyfox.com the subdomain is acme.
• API Key: Found in HappyFox under Account Settings then Developers then API.
• Auth Code: Generated alongside the API Key. Both are required.
• Brand Label: The label applied to every imported ticket. Creates or matches an existing Inbox Label.

1. Go to AI Inbox then Settings then Integrations.
2. Click Configure on the HappyFox card.
3. Enter your subdomain, API Key, Auth Code, and brand label, then click Add Account.
4. Click Run Import. The result shows how many tickets were imported and how many were skipped.
5. If more than 200 tickets are available, run the import again to continue.

Intercom

Import open conversations from Intercom into AI Inbox as Assists, including full conversation history. Use this when migrating from Intercom so your team continues working without losing any context.

Each import run brings in up to 200 conversations. You can run it again to import the next batch. Conversations already imported are skipped automatically.

• Access Token: A token created in Intercom under Settings then Integrations then Developer Hub.
• Brand Label: The label applied to every imported conversation. Creates or matches an existing Inbox Label.

1. Go to AI Inbox then Settings then Integrations.
2. Click Configure on the Intercom card.
3. Enter your Access Token and brand label, then click Add Account.
4. Click Run Import. The result shows how many conversations were imported and how many were skipped.

Zendesk

Import open and pending tickets from Zendesk into AI Inbox as Assists, including full conversation history. Use this when migrating from Zendesk so your team continues working from one workspace.

Each import run brings in up to 200 tickets. Tickets already imported are skipped automatically, so running it again is safe.

• Subdomain: The first part of your Zendesk URL. For acme.zendesk.com the subdomain is acme.
• Agent Email: The email address of the Zendesk agent used to authenticate.
• API Token: A token created in Zendesk Admin under Apps and Integrations then APIs then Zendesk API.
• Brand Label: The label applied to every imported ticket. Creates or matches an existing Inbox Label.

1. Go to AI Inbox then Settings then Integrations.
2. Click Configure on the Zendesk card.
3. Enter your subdomain, agent email, API token, and brand label, then click Add Account.
4. Click Run Import. The result shows how many tickets were imported.

Freshdesk

Import open and pending tickets from Freshdesk into AI Inbox as Assists, including full conversation history. Use this when migrating from Freshdesk so your team continues handling requests from one workspace.

Each import run brings in up to 300 tickets using Freshdesk's search API. Tickets already imported are skipped automatically.

• Domain: The first part of your Freshdesk URL. For acme.freshdesk.com the domain is acme.
• API Key: Found in Freshdesk under Profile Settings then Your API Key.
• Brand Label: The label applied to every imported ticket. Creates or matches an existing Inbox Label.

1. Go to AI Inbox then Settings then Integrations.
2. Click Configure on the Freshdesk card.
3. Enter your domain, API key, and brand label, then click Add Account.
4. Click Run Import. The result shows how many tickets were imported.

Help Scout

Import open conversations from Help Scout into AI Inbox as Assists, including full conversation history. Use this when migrating from Help Scout so your team continues working from one workspace without losing any context.

Each import run brings in up to 200 conversations. You can run it again to import the next batch. Conversations already imported are skipped automatically, so running it twice is safe.

• App Client ID: The Client ID of a Help Scout app created under Your Profile then Authentication then Apps.
• App Client Secret: Generated alongside the App Client ID. Both are required.
• Brand Label: The label applied to every imported conversation. Creates or matches an existing Inbox Label.

1. Go to AI Inbox then Settings then Integrations.
2. Click Configure on the Help Scout card.
3. Enter your App Client ID, App Client Secret, and brand label, then click Add Account.
4. Click Run Import. The result shows how many conversations were imported and how many were skipped.
5. If more than 200 conversations are available, run the import again to continue.