LaunchedEditorial Listing

Delight Customer Agent

Delight.ai (Sendbird) · Delight Customer Agent: One AI Agent for Shopping, Booking, Support and Retention Across Chat, Voice, SMS, Email and WhatsApp

Open Delight Customer Agent

Delight Customer Agent is an enterprise AI agent from Delight.ai, Sendbird's customer-agent business, that handles shopping, booking, support and retention for a brand as one agent with one memory of each customer, across chat, SMS, email, voice, WhatsApp and in-app messaging. It suits large consumer and service businesses that want a single customer-facing agent instead of separate sales, booking and support bots.

PricingCustom
Setupmedium
Runs onWeb
Open sourceNo
Customer SupportAI AgentAutonomous AgentMulti-ChannelVoice AIWhatsAppSMSEcommerceShoppingSchedulingPersistent MemoryGovernanceHuman-in-the-LoopNo-CodeContact CenterHelp DeskMCPEnterprise

Best for

Enterprise consumer and service brands (retail, travel, home services, healthcare, subscriptions) that want one AI agent to sell, book, support and retain customers across chat, voice and messaging, with one memory per customer and approval controls the CX team owns

Not ideal for

Small teams that want a self-serve support bot with public pricing, developers who expect open API documentation from day one, and companies that only need FAQ deflection inside one help desk

Who it's for

CX, digital and operations leaders at enterprise consumer and service businesses, including retail, travel, home services, healthcare and subscription brands

Capabilities

  • One customer-facing agent packaged as Delight for Shopping, Booking, Support and Retention, all working from the same customer history
  • Agent Memory Platform (AMP) that builds a per-customer profile of identity, preferences, purchase history and intent and carries it into every later conversation
  • Omnipresence: the same conversation thread across chat, SMS, email, voice, WhatsApp and in-app messaging on web and mobile, plus proactive outreach such as reminders and follow-ups
  • Agent Steward, an autonomous agent that takes over when an order or request runs into trouble, coordinating your systems and partners across sessions and following up on the customer's channel until it is back on track
  • Actionbooks: team-written, approved step-by-step flows for each kind of conversation (a return, a booking, a promotion) that set the rules, the tools the agent can call and the handoff points
  • Trust OS governance: sets which actions run autonomously and which need human approval, logs every decision and tool call, scores conversations, detects off-policy or factually wrong answers, and can pause the AI instantly
  • Release controls for agent changes: development, staging and production environments, version history with rollback, gradual rollout to a share of traffic and A/B tests
  • Midas, an AI agent for business teams that builds or changes Actionbooks from a plain-language request, shows them as a flowchart, tests them against simulated customers and holds them for approval
  • Voice agent that answers and places calls on your own numbers through Twilio, Telnyx and carrier partners, navigates mapped phone menus, masks keypad entries, and hands callers to Genesys, NICE CXone, Amazon Connect, Five9 or Cisco Finesse queues with a transcript and summary
  • Delight Desk, a built-in help desk where human agents pick up AI handoffs, including live voice transfers, in one queue for chat, email and voice; it is marketed as free with no seat fees, charging only for successful AI resolutions
  • Delight Agent MCP server for querying live agent data and updating Actionbooks, tools, knowledge sources and agent configuration from Claude, ChatGPT, Cursor or Codex
  • Commerce actions such as looking up Shopify orders, checking fulfillment status and updating shipping addresses, plus Recharge, ShipBob and ShipStation lookups
  • Enterprise security controls: role-based access, audit logs, encryption with TLS in transit, regional hosting for data residency and configurable data retention

Limitations

  • No published pricing or self-serve trial; Delight sells through a demo and an enterprise contract with onboarding by its own team
  • Developer documentation for the AI agent is on a private site that requires contacting Sendbird for access
  • Midas is rolling out to customers in stages, so access to the no-code build and simulation tool is not guaranteed at signup
  • Custom APIs and internal systems are connected with Delight's Forward Deployment team rather than fully self-serve; Intercom is listed as coming soon on the integrations page
  • Delight does not publish which foundation models currently power the agent; it describes assigning different models to different tasks rather than relying on one
  • No self-hosted deployment is documented; data residency is offered through regional hosting, but the regions are not listed publicly
  • Delight Customer Agent and Midas launched on November 9, 2026, so independent evidence on the unified four-solution setup is still limited

Use cases

  • Answering product questions, comparing options and recovering stalled carts in chat, then handling the same customer's delivery question by SMS a week later without re-explaining the order
  • Booking, rescheduling and reminding customers about appointments or service jobs over the phone and by text, including after-hours calls
  • Resolving order, return and account issues across channels and escalating to a human in Delight Desk, Salesforce or Zendesk with a summary
  • Running renewal, cancellation-save and win-back conversations for a membership or subscription business with the customer's history in hand
  • Letting a CX or operations lead change a refund policy or promotion by describing it to Midas, reviewing the flowchart and test results, and approving the release
  • Requiring human approval for high-stakes actions such as refunds above a threshold while routine requests stay automated

Our take

Delight's pitch is consolidation: one agent, one memory and one governance layer across the sales and service moments that most companies split between separate bots and vendors. That matters most for businesses where the same customer moves between buying, booking and support often, such as retail memberships, travel and home services, and where voice still carries a large share of revenue. The parts with the longest public release history are the ones inherited from Sendbird's earlier AI agent work, such as channels, Trust OS testing and rollout controls, and contact-center handoff; the unified Customer Agent packaging and Midas are new as of November 2026, and Midas is still rolling out. Because pricing, developer docs, model choices and hosting regions are not public, treat the demo as the start of due diligence and run a pilot on one or two Actionbooks with real integrations before committing to the whole lifecycle.

Who should use it

CX and digital leaders at enterprise consumer or service brands that want one AI agent to handle sales, bookings, support and retention across chat, voice and messaging, with persistent customer memory and approval controls their own team can operate.

Who should skip it

Small businesses that want a low-cost self-serve bot, developer teams that need open API docs and transparent pricing before talking to sales, and support teams that only need FAQ deflection inside their existing help desk.

Strengths

  • Covers pre-sale, booking, support and retention with one agent and one customer memory, instead of separate bots per department
  • Voice is a first-class channel with outbound calls, phone-menu navigation, keypad masking and handoff into major contact-center platforms
  • Actionbooks and Trust OS give the business explicit control over which actions need human approval, with staging, rollback, gradual rollout and A/B testing
  • Midas lets non-engineers describe changes and see them tested against simulated customers before approving them
  • Includes its own help desk (Delight Desk) but also hands off to Salesforce, Zendesk, ServiceNow, Freshworks and Sprinklr
  • Built on Sendbird's messaging infrastructure, with SOC 2 Type 2, HIPAA, GDPR and CCPA listed on its security page and Sendbird's ISO/IEC 42001 certification for AI management (announced September 2025)

Weaknesses

  • No public pricing, free trial or self-serve signup for Customer Agent
  • AI agent developer documentation is private
  • Midas is in a staged rollout
  • The unified Customer Agent packaging is new as of November 2026
  • Current model choices and hosting regions are not published

Delight Customer Agent pricing

Enterprise contract

Custom

  • Quoted after a demo
  • Dedicated onboarding and Forward Deployment team for integrations

Free tier limits: No free plan or self-serve trial for Customer Agent. Delight Desk, the help desk, is marketed as free with no seat fees; AI resolutions are billed at an unpublished rate.

Note: Delight does not publish Customer Agent prices. Its comparison pages describe contract-based enterprise pricing with dedicated onboarding, and its AWS Marketplace landing page refers to consumption-based pricing. Delight Desk, its help desk, is marketed as free with no seat fees, tier limits or annual commitment, and charges only when the AI agent resolves a conversation without escalating; the rate is not published. Sendbird's public pricing page covers its chat API, not the AI agent.

Technical specs

Modalities

Text, Voice

Where Delight Customer Agent excels

Membership retail with shopping and service in one assistant

BJ's Wholesale Club says members talk to the same agent whether they are browsing for a grill, checking club hours or asking about a renewal, which is the cross-lifecycle setup Customer Agent is built for.

After-hours calls for a home services business

The voice agent answers on the business's own numbers, qualifies the caller and books the job, and calls back with context if the line drops.

Policy changes owned by the CX team

Midas turns a plain-language change into an updated Actionbook, tests it against simulated customers and waits for approval, so ops teams can adjust offers and policies without an engineering ticket.

High-stakes actions with human sign-off

Trust OS lets the business automate routine requests while sending refunds, exceptions or ambiguous cases to a person with full context.

Delight Customer Agent vs. competitors

Delight Customer Agent vs. Sierra

Sierra is an enterprise platform for branded agents that handle support, sales and account workflows, sold on custom contracts. Delight competes for the same buyers, emphasizing a single per-customer memory across shopping, booking, support and retention, its own help desk, and Midas for building and testing changes in natural language.

Delight Customer Agent vs. Decagon

Decagon is an enterprise AI support platform for large consumer brands across chat, voice, email and SMS with shared context. Delight covers similar channels but frames the agent around the whole lifecycle, including shopping, booking and retention, and bundles Delight Desk for human handoff.

Delight Customer Agent vs. Ada

Ada is a standalone AI customer service platform with agents across web, mobile, email, voice, SMS and social messaging. Delight is also help-desk-independent, but adds sales and booking flows, persistent per-customer memory and its own help desk, and is sold only on enterprise contracts.

Delight Customer Agent vs. Fin

Fin is an AI support agent that runs on Intercom and other help desks with a published per-outcome price and a self-serve trial. Delight is aimed at enterprises that want one agent across sales, booking, support and retention, with voice and contact-center handoff, sold through a demo and a contract.

Frequently asked questions

What is Delight Customer Agent?

Delight Customer Agent is an AI agent from Delight.ai that serves a customer across the whole relationship: finding and buying a product, booking an appointment, getting support and renewing or coming back. It is sold as four solutions (Delight for Shopping, Booking, Support and Retention) that share one customer history through Delight's Agent Memory Platform, and it works across chat, SMS, email, voice, WhatsApp and in-app messaging.

Is Delight.ai the same company as Sendbird?

Delight.ai is Sendbird's customer-agent business. The launch release describes it as Sendbird's portfolio of customer agents for enterprises, the delight.ai site is published by Sendbird, Inc., and product updates released in 2025 under the "Sendbird AI agent" name now appear as Delight.ai releases. Sendbird's chat, voice and messaging APIs continue under the Sendbird brand.

How much does Delight Customer Agent cost?

Delight does not publish prices for Customer Agent. Its comparison pages describe a single enterprise contract with dedicated onboarding, and you get a quote through a demo. Its Delight Desk help desk is marketed as free, with no seat fees, tier limits or annual commitment; you pay only when the AI agent resolves a conversation without escalating, and that rate is not published.

How does Delight decide what the agent can do without a human?

Each task runs on an Actionbook, a step-by-step flow your team writes and approves. Trust OS, Delight's governance layer, sets which actions the agent can take on its own and which need human approval, logs every decision and tool call, and can pause the AI with one switch. Changes go through development and staging, can be rolled out to a share of traffic or A/B tested, and any version can be reverted.

What is Midas?

Midas is an AI agent in Delight.ai for the team that runs the Customer Agent. You describe what the agent should do in plain language; Midas writes it into an Actionbook, shows it as a flowchart, asks about gaps in your policies, tests the change against simulated customers and sends it for approval. It also reviews conversations and proposes the next improvement. Midas does not change the live agent on its own, and Delight says it is rolling out to customers in stages.

Does Delight work with Salesforce, Zendesk or my contact center?

Yes. Delight's integrations page lists Salesforce, Zendesk, ServiceNow, Freshworks and Sprinklr for handoff and knowledge, and Genesys, NICE CXone, Amazon Connect, Five9, Cisco Finesse, RingCentral and Zoom Phone for routing voice calls or conversations to human agents with context. It also has its own help desk, Delight Desk. Other systems can be connected through APIs and webhooks with Delight's Forward Deployment team.

Delight vs Sierra: what is the difference?

Both are enterprise platforms for branded, action-taking customer agents sold on custom contracts. Delight positions its agent around one persistent memory per customer shared across shopping, booking, support and retention, ships its own help desk, and gives business teams Midas to build and test changes in natural language. Sierra is an enterprise platform for branded agents that handle support, sales and account workflows in the company's own voice; pilot both on the same workflows and integrations before choosing.

Integrations & fit

SalesforceZendeskServiceNowFreshworksSprinklrDelight DeskGenesysNICE CXoneAmazon ConnectFive9Cisco FinesseRingCentralZoom PhoneTwilioTelnyxWhatsAppShopifyRechargeShipBobShipStationConstructor.ioNavitaireHousecall ProOptixAuditDataConfluenceNotionGoogle DriveAmazon S3ZapierMCP
Good fit forEnterprise
Pricing modelCustom· Contact for pricing
See pricing on Delight Customer Agent →

Alternatives to consider

About Delight Customer Agent

Delight.ai is the customer-agent business of Sendbird, the in-app chat and messaging infrastructure company, which the launch release calls its parent company; the delight.ai site is published by Sendbird, Inc., and features released during 2025 as the "Sendbird AI agent" now appear as Delight.ai releases. Delight Customer Agent, launched on November 9, 2026, packages that platform as one agent for the whole customer relationship. It is sold as four solutions on the same foundation: Delight for Shopping (product questions, comparisons, add-ons, cart recovery and checkout in the conversation), Delight for Booking (qualifying inquiries, booking appointments and jobs, reminders and reschedules), Delight for Support (order, return and account issues with escalation) and Delight for Retention (renewals, cancellation saves and win-backs). What makes it one agent rather than four is the Agent Memory Platform (AMP), which keeps a profile of each customer's identity, preferences and intent that every conversation reads and adds to, and Omnipresence, which carries that thread across channels so a conversation that starts in chat can continue by SMS or phone. Agent Steward, an autonomous agent, takes over when an order or request runs into trouble, coordinating systems and partners and following up until it is resolved. Tasks run on Actionbooks: step-by-step flows that the business writes and approves, holding the rules, the tools the agent may call and the points where it hands off to a person in Delight Desk (Delight's own help desk) or an existing help desk. Trust OS is the governance layer around this: it decides which actions the agent may take alone and which need human approval, logs every decision and tool call, scores conversations, and moves changes through development, staging, gradual rollout or A/B tests, with versions that can be reverted and a switch that pauses the AI. Midas, launched the same day, is an AI agent for the business team rather than the customer: you describe a change in plain language, Midas writes it into an Actionbook, draws it as a flowchart, tests it against simulated customers, and nothing goes live until someone with approval rights publishes it; Delight says Midas is rolling out to customers in stages. Integrations cover help desks and CRMs (Salesforce, Zendesk, ServiceNow, Freshworks, Sprinklr), contact centers for voice handoff (Genesys, NICE CXone, Amazon Connect, Five9, Cisco Finesse, RingCentral), knowledge sources (Confluence, Notion, Google Drive, Amazon S3) and commerce and industry systems (Shopify, Recharge, ShipBob, ShipStation, Constructor.io, Navitaire, Housecall Pro), with custom APIs and webhooks wired in by Delight's Forward Deployment team. The tradeoffs: Customer Agent is sold through sales on an enterprise contract with no published price or self-serve trial (only the Delight Desk help desk is marketed as free, with charges per AI resolution at an unpublished rate), the developer documentation sits on a private site, and the claim that one agent beats specialized tools depends heavily on how much integration and Actionbook work your team and Delight's put in.

Updates from Delight Customer Agent

LaunchDelight.ai launches Delight Customer Agent and Midas

Delight.ai introduced Delight Customer Agent, one AI agent covering shopping, booking, support and retention on a shared customer history, and Midas, an AI agent that lets business teams build, test against simulated customers, improve and analyze their Customer Agent in natural language, with changes going live only after team approval. Delight says Midas is rolling out to customers in stages.

IntegrationDelight Agent MCP released

Delight released an MCP server that connects Claude Code, Claude.ai, ChatGPT, Cursor and Codex to live agent data, so teams can query resolution rates, transcripts and quality signals and create or update Actionbooks, tools, knowledge sources and agent configuration from their AI client.

Are you the founder? Claim this listing →