chatAgent

An AI shopping assistant for WhatsApp — Rivet durable actors, AgentOS V8 isolates, Pi Agent Core, Shopify Storefront, and Jev turn routing.

chatAgent answers customer messages for a Shopify store on WhatsApp. Each customer gets a durable Rivet actor with its own AgentOS V8 isolate, where a Pi Agent Core agent calls the Shopify Storefront API and replies in the store’s voice.

A Jev (TypeSafe) judgment runs before every turn: it decides whether the message needs the full shopping agent or can be answered directly, and it scores the catalog keywords the search tool should use.

Start here

WhatsApp Shopping Bot

What the bot does, end to end.

Architecture

The layers: gateway, actors, isolates, agent runtime.

Rivet Actor Model

The actor implementation, canonical reference.

Turn Router Middleware

Jev routing, the fast lane, and the scored search keywords.

Run it

bun install                 # repository root, or any package folder
bun --cwd runtime run dev   # the runtime + actors
bun test                    # runtime unit and integration tests

The service runs on Node 22 with RivetKit and AgentOS; Bun is the package manager, task runner, and test runner. Configuration lives in .env — see the Integration Guide for the full wiring and Concurrency & Security for the limits and guards.

How this site is built

These pages are the repository’s own docs/ folder, published by Blume and deployed to Cloudflare Pages on every push to main — see Docs Site for the pipeline, the secrets it needs, and the Cloudflare Access layer in front of it.