Answers your customers can verify.

Turn your documents and web pages into a support chatbot that answers from your knowledge, cites the source, and knows when it does not have enough information.

Grounded answersPDF and textUp to 15 web pages
Relay workspaceKnowledge ready
Website support botLive preview
CustomerCan I return an opened item?
Relay

Opened items can be returned within 14 days when they are complete and undamaged.

Returns policy.pdf
Ask from connected knowledge

From source to published chatbotIngestRetrieveAnswerCitePublish

A useful answer is only half the job.

Relay keeps the supporting evidence close enough to inspect. Small teams get a practical support tool; visitors can see why an answer deserves trust.

What does the warranty cover?
Relay

The warranty covers manufacturing defects for 12 months from the purchase date. Accidental damage is excluded.

Warranty guide.pdf
01

Evidence stays visible

Every grounded response can name the document or webpage that supported it.

02

Weak matches stop safely

When retrieval confidence is poor, Relay uses the configured fallback instead of unrelated context.

03

Publish where people need it

Share a hosted page or add the chatbot to a website with an embed.

Build the knowledge once. Test before you publish.

The workflow keeps setup understandable for a small team without hiding what is happening to its content.

Connect

Add useful knowledge

Upload a text-based PDF, paste text, or scrape a public HTTPS website.

Service handbook.pdfUploading · 72%
Verify

Ask real questions

Test the chatbot and check the cited source before anyone else sees it.

Answer grounded2 relevant chunks retrieved
Publish

Choose the right surface

Open a public share page or copy the website embed when the answers are ready.

Share pageCopy embed

Built as a real full-stack product.

Relay is also a portfolio project by Mushraf Moosa Shaikh. The interface, API, retrieval pipeline, storage, authentication, and production safeguards are implemented as separate, inspectable layers.

01ClientNext.js, React, Clerk
02APIExpress, Zod, rate limits
03RetrievalEmbeddings, PostgreSQL, pgvector
04DeliveryPrivate files, share page, embed

Know the boundaries before you connect a source.

Which files work?

Relay accepts text-based PDF and TXT content. Scanned PDFs need text recognition before indexing.

How much can a website scrape?

Choose 5, 10, or 15 pages for a public HTTPS website.

Where is retrieval stored?

Knowledge chunks and embeddings are stored in PostgreSQL with pgvector.

What happens without a strong match?

Relay returns the chatbot’s configured fallback instead of passing unrelated context to the model.