Jhunkoo.ai

Jhunkoo.ai is our own product: a platform for building AI agents that answer website visitors' questions and capture leads, trained on a business's own website, documents and APIs.

Businesses like what ChatGPT can do, but it doesn't know their business. An agent that does — trained on their own content and connected to their own systems to book meetings and capture leads — has meant hiring developers most businesses don't have the resources for.

Take a business from its own content to a live agent on its website without writing code — define the agent, add the data, connect integrations, go live.

Multi-tenant from day one: every agent belongs to an organisation, with its own team and access, or to a personal workspace. Retrieval, queues and rate limits run on managed services, so the platform scales without servers to operate. The widget installs with a single script — floating or inline, loaded up front or lazily — or through Google Tag Manager or Segment, with guides for React and Next.js, Vue, Nuxt, Astro and Svelte. Models stay swappable across providers, with usage and cost visible to the customer throughout.

A Next.js application on Vercel behind Cloudflare, with PostgreSQL through Prisma, sign-in and organisations through Clerk, and files in Cloudflare R2. Pinecone holds the vector index behind retrieval, Upstash handles Redis and rate limiting, and QStash runs background jobs. Agents run on 19 models across seven providers — OpenAI, Anthropic, Google, Mistral, xAI, DeepSeek and Moonshot AI — through the Vercel AI Gateway, with tokens and cost recorded for every message. The documentation is built with Nextra.

• Next.js
• Vercel
• Cloudflare
• PostgreSQL
• Clerk
• Pinecone
• Upstash
• QStash
• Vercel AI Gateway
• Nextra

Screens captured from the live sites.