AIKON

An in-app AI agent for a wholesale electronics marketplace. It creates and finds offers through conversation and answers questions from the client's own documents.

AIKON is a trading floor for wholesale electronics, where buyers and sellers connect through offers. The matching already worked, but users had to learn the interface to use it. AIKON wanted an assistant that does the work in conversation instead.

We added an agent inside the existing product without disrupting the flows current users relied on. It creates an offer from a plain description of what the user wants, finds matching offers instead of leaving the user to filter by hand, and manages the user's contacts and lists on request.

It answers questions about the platform from a knowledge base the admins maintain themselves. Retrieval-augmented generation grounds every answer in AIKON's own documents, and the admins update that knowledge without a deploy. The agent remembers earlier conversations, picks up where they left off, and reads the user's profile to keep the conversation relevant.

The agent runs on LangChain over OpenAI, so the model is a configuration choice rather than a rewrite, with a Node.js backend. Embeddings live in PostgreSQL with pgvector, next to the app's own data, so there is no second database to run. Langfuse traces every call, so the client can see what each conversation did and what it cost.

The agent was delivered in July 2026, three weeks from start to handover, into a product that already had users. We publish no adoption figure for it. The agent is live in the product today.