Project · Own product

A chatbot that answers with your information and shows where it came from.

Two platforms on the same foundation: DocuBot builds a chatbot from your documents and Sitebot from your website. Both index the content, answer citing their sources and give you a snippet to add the chat to any page.

What it does

Key features.

Generic chatbots make things up. To be useful in a company, an assistant has to answer with that company's information (manuals, policies, catalogs or website content) and show where the answer came from.

01

DocuBot: your documents

Upload files or whole folders: PDF, Word, Excel, CSV, TXT and Markdown.

02

Sitebot: your website

Crawls the site via robots.txt, the sitemap or internal links, and turns each page into text.

03

Answers with sources

Every answer shows which document or page the information came from.

04

Embeddable chat

A snippet to add the chat to any site, restricted to authorized domains.

05

Incremental updates

When re-uploading or re-crawling, only what changed gets reprocessed.

06

Management dashboard

Collections or sites, settings and limits from a dashboard with login.

Under the hood

The technical challenges.

RAG on pgvector

Semantic search in PostgreSQL with pgvector: the model answers only from the most relevant chunks.

Local embeddings

Multilingual embeddings are computed on the server itself, with no dependency on an external provider.

Background processing

Each file is processed in a queue (BullMQ) and answers are streamed.

Security

JWT authentication, widget origin validation, rate limiting and SSRF protection in the crawler.

Stack

What it's built with.

NestJSNext.jsTypeScriptPostgreSQLpgvectorDrizzle ORMBullMQPlaywrightDocker
Type
Own product
Status
Own product, working.
Website
—
Let's talk

Need something similar?

Tell me your idea or how your company runs. I'll reply personally within 2 business days with next steps.

Go to the contact form →