Skip to main content

Customer Support

Lira's Customer Support module gives your organisation a fully autonomous AI support operation — handling inbound emails, in-app support, live chat, voice calls, tickets, and AI actions, all grounded in your Knowledge Base.

How Lira decides what to say

Lira uses updated frontier language models from the Claude and GPT families, routed through a single provider-agnostic adapter. The adapter picks the best provider per organisation and automatically falls back to a secondary provider if the primary fails before the first token — so a single API outage never leaves a visitor staring at a half-formed reply.

For every visitor message, the Organization Context System assembles a fresh context bundle from four sources — your organisation profile, retrieved knowledge-base chunks, live product context from the SDK, and recent conversation history — before the model composes the reply. That's what makes the answers feel specific to your org and to the conversation in progress, not generic AI fluff. See Organization Context System for the full assembly order.

Once activated, Lira reads every incoming message, searches your documentation for the best answer, and responds confidently. When it can't, it opens a ticket for your team to handle asynchronously — without breaking the live chat with the customer.

What's included​

ModuleWhat it does
TicketsAsync human-followup queue. The primary surface your team works from.
Web SDKFull-page support embed for your own /support route, plus identity and live context
Chat WidgetEmbeddable floating chat button for your website
Hosted PortalOptional no-code fallback page when you cannot ship the SDK yet
Agent RuntimeThe capability model, risk tiers, and policy engine that decide what the AI is allowed to run
CapabilitiesAdmin catalog of resources and actions the AI can call — override risk, scope, and description
ActionsThe human-approval queue for runs the policy engine routed to a teammate
AuditEvery action run the agent made, with policy decision, redacted inputs and outputs, and estimated cost
ProactiveAutomated outreach — trigger messages based on customer events
AnalyticsCSAT scores, resolution rates, response times, and weekly reports
Chat historyRead-only audit log of every AI chat. Use it to QA Lira's accuracy.
SettingsConfigure channels, behaviour, ticketing email, and widget appearance

How it works​

When support is activated, Lira listens across every channel you've enabled:

Customer message (email / SDK / chat / voice / hosted portal)
  → Lira retrieves relevant content from your Knowledge Base
  → Decides whether to read product data or perform an action
  → Policy engine checks the capability's risk + the visitor's auth scope
      ├─  auto-execute    →  run the capability now
      ├─  customer confirm →  ask in chat, then run
      ├─  human approve    →  queue in Actions, wait for a teammate
      └─  block            →  refuse, escalate or hand off
  → Generates a grounded response with whatever facts were collected
  → Confidence ≥ threshold  →  sends reply autonomously
  → Confidence   < threshold  →  opens a Ticket — async human follow-up
                                  (the live chat keeps going in parallel)

Every capability call is recorded as an action run with the policy decision, redacted inputs and outputs, and estimated model cost. See Agent Runtime for the full model, Capabilities to manage the catalog, and Audit to read the trail.

The AI chat is never interrupted. When something needs a human, Lira opens a Ticket and keeps chatting with the visitor. Your team works the tickets queue at their own pace; the visitor gets an email when there's a reply.

Every raw chat — across channels — is still archived in Chat history so you can audit Lira's accuracy.

Channels​

Lira supports several support surfaces, each independently configurable from Support → Settings:

Email​

Lira is assigned a platform support address the moment you activate (e.g. [email protected]). You can share this directly with customers, or configure your own address (e.g. [email protected]) and set up a forwarding rule — Lira handles everything from there.

Chat Widget​

A floating chat button embedded on any page of your website. Install it with a single <script> tag. Customers get instant AI responses without leaving your site. Fully customisable colour and greeting message.

Web SDK​

The recommended B2B integration. Create your own support route, for example lemonpay.com/support, and mount Lira inside it. Your company owns the domain, app shell, and surrounding UI; Lira powers the AI conversation, tickets, signed identity, live product context, and action workflow. → Full SDK guide

Voice​

Inbound phone support powered by Lira's real-time voice AI. Customers call your support line; Lira answers, understands their issue, and either resolves it or escalates to a human in the same workflow.

Hosted Portal​

A branded, publicly accessible fallback page at support.liraintelligence.com/your-slug. Use it for temporary no-code launches or email links when a customer cannot integrate the Web SDK yet. → Hosted portal guide

Getting started​

If you haven't activated the support module yet, the app will guide you through a short setup wizard covering email, channels, integrations, and knowledge base seeding.

→ Activation guide