RAG / production-readylive

Deploy a custom AI trained on your website in minutes.

LargeChat crawls your site, follows important files, builds a searchable knowledge base, and answers with citations your users can verify.

Hybrid retrieval Cited answers Auto branding
beaconview.test bot
You
Ask anything from the trained site...
Enter
Built for sites where details matter
Beaconview PrepCobalt CreekJuniper CartMossvale DentalLumen LedgerHarbor Finch LawPine & ParcelNova Nest HOA

The machinery behind a bot that works out of the box.

The design is quiet, but the pipeline is not. Crawl coverage, retrieval scoring, branding, and citations all work together.

01

Website training that keeps digging

Crawl pages, PDFs, calendars, linked files, image-only buttons, and common sitemap patterns without asking customers to configure every edge case.

> crawl.website({ depth: "adaptive" })
02

PDFs and pages in one answer

Uniform policies, handbooks, calendars, agendas, and ordinary web pages are chunked into one retrievable knowledge base.

> source: beaconview-policy.pdf + /student-life/uniforms
03

Hybrid retrieval by default

Semantic search, keyword matches, title boosts, FAQ matches, recency, and source priority combine before the answer is generated.

> semantic + keyword + rerank
04

Strict citation mode

Answers cite the source blocks they came from and fall back when the retrieved context is not strong enough.

> min_score: 0.62
05

Debuggable from day one

See retrieved chunks, component scores, fallback reasons, crawl errors, and unanswered questions in a single product workflow.

> retrieval.playground.open()
06

Auto-branded widgets

Pull favicon, logo, theme color, and chat palette from the source website so new bots look native out of the box.

> branding.detect(siteUrl)

Three steps. No services team required.

first answer in minutes
01

Create

Enter a website URL. LargeChat creates the bot, detects branding, and prepares the crawler.

  • Logo and colors
  • Default prompt
  • Domain-ready widget
02

Train

The crawler collects the useful pages and files, chunks content, embeds it, and records training diagnostics.

  • Pages, PDFs, events
  • Duplicate handling
  • Structured facts
03

Answer

The bot retrieves the best context, answers with citations, and shows what was used when you debug it.

  • Hybrid retrieval
  • Cited responses
  • Fallback when unsure

Every answer, cited.

Inline source pills make the answer auditable. The retrieval debugger shows the exact chunks and component scores behind each response.

retrieval.config6 chunks / 5.8k tokens
semantic_top_k: 60
keyword_top_k: 40
min_score: 0.62
citation_mode: strict
User asked
What are the uniform rules for field trip days?
LargeChat replied1.2s / 241 tokens
Beaconview Prep's Uniforms page says field trip attire follows the standard daily uniform unless the event notice says otherwise 1student-life/uniforms. The linked policy PDF lists the required polo, navy bottoms, closed-toe shoes, and outerwear rules 2beaconview-dress-policy.pdf.
Retrieved sources
1student-life/uniforms0.91
2beaconview-dress-policy.pdf0.88
3families/pto-board0.83
4calendar/spring-fair.json0.78

Start simple. Scale when traffic arrives.

Plans are based on bots, message credits, and training capacity.

Free

01

For testing a single bot.

$0
Start free
  • 1 bot
  • 50 message credits
  • 10 training links
  • Embed on unlimited sites
  • LargeChat branding
Most teams

Hobby

02

For a real website with steady traffic.

$40/ mo
Start building
  • 1 bot
  • 1,500 message credits
  • Unlimited training links
  • Advanced models
  • Basic analytics
  • API access

Enterprise

03

For teams that need limits, controls, and support.

Custom
Contact us
  • Higher limits
  • Priority support
  • SLA options
  • Custom integrations
  • Dedicated account support

Questions, answered.

LargeChat centers on a clean RAG workflow: train, retrieve, answer, inspect.

Start with one source-grounded bot. Add sources, tune retrieval settings, and adjust the prompt as you learn what users ask.

Training means crawling, parsing, cleaning, chunking, embedding, indexing, and testing retrieval. The LLM is not fine-tuned for ordinary website knowledge.

Yes. Linked PDFs and website pages are indexed together, so a question can be answered from the most specific source available.

The crawler extracts useful alt text, image filenames, and linked pages, then follows important image-only links such as app announcements or policy buttons.

Use the retrieval playground and training diagnostics to inspect crawl errors, skipped pages, retrieval scores, and the final context sent to the model.

Ship the bot you would build yourself, in an afternoon.

No credit card / one source-grounded assistant
LargeChatdemo
LargeChat
Ask about training, citations, PDF retrieval, or website branding.