Skip to main content

The platform

Retrieval, knowledge, widget and reporting — engineered into one product your agency puts its own name on.

Nothing is rented from a chatbot vendor and rebadged. What follows is the whole feature set, as it actually works.

Self-contained engine

Nothing sits underneath.

Most "white-label chatbots" are a settings page on top of someone else's engine — a vendor you quietly rent, whose name and limits leak into your margin. Replica's retrieval engine is part of the product: each bot runs its own model against its own client's documents, answers stream token by token, and every reply names the source it retrieved from.

  • Per-bot model — each bot retrieves from one client's knowledge base only; answers never bleed across accounts.
  • Streamed answers — replies type out live instead of arriving as a wall of text after a spinner.
  • Retrieval with a source — every answer cites the document it came from, so a wrong answer is traceable to a wrong document.
Anatomy of one answer
VISITOR QUESTIONDo you ship to PO boxes?PER-BOT RETRIEVAL MODELNorthlake Dental's bot onlysearches 1,847 chunks from this client's 116 documentsSTREAMED ANSWERYes — standard post, three to five business days…source: FAQ — Billing.txt · confidence 0.87deflected

No vendor in the middle: question, model and answer all live inside your client's account.

The ingest pipeline — every state, named
  1. queued

    The file or URL is accepted and waits its turn — nothing is silently dropped.

  2. parsing

    Text is extracted from PDF, TXT, DOCX or the crawled page.

  3. chunking

    The text splits into retrievable chunks — from 312 to 1,847 per bot in practice.

  4. embedding

    Each chunk is indexed into the bot's retrieval model.

  5. ready

    The document is live in the bot's brain and citable in answers.

failed

…and when ingestion can't complete, the document says so — with the reason attached. "Brochure scan.pdf — failed: no text layer. Re-upload as PDF with text, or as TXT." No silent gaps in what the bot knows.

Knowledge-base management

An honest pipeline beats a magic one.

Upload a PDF, TXT or DOCX, or point the crawler at a URL. Ingestion is asynchronous and says so: you always see which documents are live in the bot's answers and which are still working through the queue. A technical buyer can watch the states change; a client just sees a bot that knows their business.

  • Four source types — PDF, TXT, DOCX and URLs, mixed freely in one knowledge base.
  • Re-index on demand — replace a document and the bot's answers follow the new version.
  • Plan-scale capacity — from 24 documents on Starter to 480 on Scale.

Embeddable widget

One script tag. Every client's own brand.

The widget drops into any client's website with a single line of code and runs in an isolated frame, cross-domain. Colours, wordmark, greeting and position are configured per client — and the widget only answers on domains you've whitelisted, so a copied snippet is useless anywhere else.

  • Cross-domain and isolated — its styles can't leak into the host site, and the host can't break it.
  • Branded per client — Northlake Dental's widget and Harbor & Finch's widget share an engine and nothing else.
  • Domain-whitelisted — the snippet answers only on the client's registered domains.
Try the live demo
Same engine, two clients
NDNorthlake Dental — front-desk assistant
whitelisted: northlakedental.comtheir colours, their greeting
HFHarbor & Finch Law — intake screener
whitelisted: harborfinchlaw.comtheir colours, their greeting

The visitor on either site never sees Replica — or your agency's other clients.

The differentiator

The deflection & gap report. No competitor ships this.

Every period, every client gets three things in one artifact: the deflection rate their bot achieved, the support cost that avoided, and — the part no other white-label chatbot platform produces — the exact questions the bot answered poorly, clustered by topic with confidence scores. It is the proof of value your agency bills against, and it is included on every plan.

Deflection & gap report

Northlake Dental · Feb 1 – Feb 28, 2026 · compared with January

74%deflected
Estimated saving
$6,200
8,930 messages · vs 68% in January

Across client bots the deflection rate runs 61–78% — here, 6,608 of 8,930 conversations closed without a human. The questions the bot fumbled cluster into the topics below, each one a document away from being deflected next month.

Typical monthly volumes range from 1,240 to 24,100 messages; estimated savings run $1,840 to $6,200 per client.

Questions the bot answered poorly — by topic
  • Insurance accepted96 questions · confidence 0.42
  • Weekend appointment availability88 questions · confidence 0.51
  • Refund window after 30 days84 questions · confidence 0.58
  • Cancelling mid-contract76 questions · confidence 0.63
the agency presenting the report to its client

The renewal meeting, pre-written.

A number the client already felt — a quieter inbox — turned into a dollar figure, plus an honest list of what to fix next. It exports per client, per period, under your agency's brand. We have not found a competing white-label platform that produces it at all.

See the widget that feeds it

Multi-client by design

One login for you. A whole platform for each client.

Your agency runs every client from one workspace. Each client account is a sealed unit: its own bots, its own knowledge base, its own branding, its own login. Client staff who sign in see your agency's name and their own data — never Replica's, never each other's.

  • Sealed accounts — no client's data, documents or conversations are visible to another.
  • Client-side logins — a client's team sees a dashboard with your brand and only their numbers.
BS
Brightline Studio
agency workspace · Pro plan · 10 client accounts
Northlake Dental
3 bots · 116 documents · own widget branding · own client login
Harbor & Finch Law
2 bots · 480 documents · own widget branding · own client login
Verdant Home Services
4 bots · 24 documents · own widget branding · own client login
+ every other client, each sealed the same way

And the rest of the job

The capabilities you assumed were there. They are.

Conversation history

Every bot conversation stored as a full transcript, searchable per client.

per bot, per client
Live handoff

A human steps into any thread mid-conversation when the bot shouldn't answer.

bot → human, in place
Usage tracking

Messages counted per client against the plan, visible before anyone hits a limit.

1,240 – 24,100 /mo by plan
Bulk import / export

Clients, documents and transcripts move in and out as CSV — no lock-in either way.

CSV, both directions
Per-client permissions

Client logins see only their own bots, transcripts and reports — nothing agency-level.

sealed by account

Run it under your name, on your first client, this week.

Engine, knowledge pipeline, widget and the gap report are included from the Starter plan up.

Get startedSee pricing