Your entire contact center, working as one.

Dual-engine routing

Routine questions run on a fast model. The moment a case turns emotional or expensive, it escalates to a heavier reasoning engine mid-conversation.

In plain English: a fast, cheap model handles routine questions. The moment a customer sounds upset or a refund is on the line, Katalyst switches to a smarter model automatically.

Real-time knowledge graph

Products, policies, promotions and resolved cases are indexed into one searchable graph and refreshed on a schedule, so every answer is grounded in your live store.

Zero-guess protocol

If confidence drops below your floor, the concierge says it will check instead of inventing an answer — and files the question into a learning queue for you to teach.

A staffed virtual call center

Chloe handles sizing, Marcus tracks orders, Devon runs returns and Sarah leads escalations — each with their own directive, tools and concession cap.

Inbound voice concierge

A phone agent with natural prosody, instant barge-in handling, caller-ID memory and full transcripts of every turn — with SMS follow-up if a caller drops mid-lookup.

Verification and one-time passcodes

Adaptive ID&V gates sensitive actions: knowledge-based checks for order lookups, OTP over SMS or email before address changes and refunds.

Deterministic escalation matrix

Explicit demands, risk keywords, repeated rejections and conversation loops trip named rules that route the case up the chain — you can see and edit every rule.

Live monitoring, barge and whisper

Watch active calls and chats in real time, whisper coaching to the agent, or take the conversation over yourself without dropping the customer.

Autonomous QA and coaching

Every interaction is graded on empathy, policy adherence, tool execution and resolution, with structured summaries and issue flags to speed up review.

Refunds on first scan

Prepaid label issued, carrier scan detected, refund released — before the parcel reaches your warehouse.

Guardrails you set, not the model

Per-agent spend and concession caps are hard limits. Katalyst can't exceed them without human approval, and every autonomous action is logged and reviewable.

Native multi-language replies

Shopper language is detected per message and answers come back in that language — across chat and voice, with no separate configuration.

Analytics, alerts and a morning digest

Resolution rate, lookup accuracy and time-to-first-response tracked per merchant, with threshold alerts and an executive digest delivered every morning.

Patterns, not tickets

Three customers reporting the same seam failure becomes one operational alert instead of three closed conversations.

Live commerce integrations

Connect Shopify, Stripe, EasyPost, Twilio and Resend with your own keys for real catalog sync, refunds, labels, SMS and email — or leave them off and run in sandbox.

Embeddable in minutes

A branded widget that inherits your colours, typography and voice — drop it on any storefront via iframe or a one-line script.

Service & appointment mode

Not a store? Switch to service mode: the concierge asks your intake questions, qualifies the lead, offers real calendar slots and books the visit — bridging true emergencies straight to you.

In plain English: home services, clinics and professional practices get a booking agent instead of an order agent, with every qualified lead in one queue.

High-volume concurrency

Parallel agents auto-scale to your line limit so nobody waits on hold. A live wallboard shows active chats, texts and calls with utilization and peak-hour metering.

Interactive hold companion

Callers parked for a specialist get a texted link to a live companion page: a running wait clock, their place in the queue, and a carousel of hold stations that changes the music on the phone line instantly.

In plain English: instead of dead hold music, the caller picks their own vibe and watches the wait tick down on their phone.

Frustration pre-screening & screen-pop

Mood check-ins, wait time, restless station skips and tap telemetry feed one weighted score. Hot callers auto-route to a supervisor, who sees a context bar, restlessness meter and an AI executive brief with the exact opening line.

In plain English: the person who picks up already knows who is upset, why, and how to open the call.

Verify-while-you-wait & smart callbacks

Order number and ZIP are verified on hold, so the specialist opens the call with the order already on screen. Or skip the queue entirely — a priority callback dials back with full context within minutes.

Spam, robocall & collections shield

Known robocallers and debt-collection scripts are screened before an agent answers, with blocklists, an audio challenge for unknown numbers, and a quarantine log of minutes saved.

Custom voice studio

Record a short sample in the browser and clone it — your concierge answers the phone in your own voice, or a polished sandbox voice until you connect a provider key.

Autopilot or co-pilot

Choose full autonomy or approval-first. Set autonomy caps and a confidence floor, and Katalyst drafts instead of acts whenever a case crosses your line.

Accuracy & guardrails studio

An accuracy scorecard, a simulate-and-audit sandbox to test edge cases before customers hit them, and plain-English feedback distilled into toggleable learned rules.