Back to portfolio
ProductIn production2025–Present

Klyro

Embeddable AI chat for any website. Upload knowledge, shape a persona, drop in a script or npm package — visitors get accurate answers from your data, not a generic bot. Built end-to-end as a dashboard + RAG engine + zero-dependency widget (@klyro/widget).

Gallery

Knowledge base

Upload docs, paste text, or scrape URLs — the bot only answers from what you approve.

Persona control

Name, tone, and boundaries so the assistant sounds like you, not a stock chatbot.

Embed anywhere

One script tag or npm install @klyro/widget — light, themable, domain-locked.

Product

Klyro is an AI chat platform built for sites that need a controlled assistant, not a black-box support widget. Owners curate knowledge, configure how the bot speaks, and embed it in minutes — with RAG so answers stay grounded in their content.

Problem

Static sites force visitors to hunt for skills, projects, and availability. Generic chatbots invent answers.

Solution

RAG over a curated knowledge base, custom persona, and a drop-in widget (script or React) with full look-and-behavior control.

Highlights

  • LLM-powered RAG with pgvector semantic search over your documents
  • Zero-dependency embeddable widget published as @klyro/widget
  • Persona presets plus custom tone, traits, and share boundaries
  • Multi-format ingestion: PDF, DOCX, TXT, MD, and URL scraping
  • Tool calling for live GitHub repos and Calendly availability
  • Admin dashboard: knowledge, widgets, test chat, and integrations
  • PWA-ready dashboard with offline support

Specs

Type
AI product / developer tool
Role
Founder · Full-stack engineer
Company
Independent
Live
klyro-pro.vercel.app
Focus
Embeddable AI chat & knowledge-grounded answers
Period
2025–Present
Package
@klyro/widget · v2.3.5 · MIT
Traction
1,100+ weekly npm downloads in the first week

How it works

  1. 1Upload knowledge — Documents, text, or URLs become the bot’s source of truth.
  2. 2Configure persona — Set name, style, and what the assistant is allowed to share.
  3. 3Embed the widget — Script tag for any site, or React via @klyro/widget.
  4. 4Visitors ask, RAG answers — Semantic search + LLM responses grounded in your data (with optional live tools).

Achievements

  • Shipped a production dashboard + chat API + npm widget as one product
  • Published @klyro/widget — 1,100+ weekly downloads shortly after launch (v2.3.5, 12 versions)
  • Zero-dependency widget (~25KB minified; MIT; script or React module)
  • Full RAG pipeline: chunking, embeddings, pgvector retrieval, strict knowledge mode
  • Integrations for GitHub project context and Calendly scheduling

Stack

TypeScriptNext.jsReactOpenAISupabasepgvectoresbuildTailwind
Core StackFrontendBackend & DatabasesCloud & DevOpsData & Machine LearningDesign & DX

Links

Live dashboard@klyro/widget