All case studies
Internal ToolsFive Star Group

The Chat App With AI Teammates in the Room

A cross-platform team chat app built to end the per-seat SaaS bill — with AI agents wired in as first-class teammates through a bot API, not bolted on through an integration marketplace.

Team ChatCross-PlatformAI-Native
The Challenge

Every mainstream team-chat product bills per seat, forever — the invoice grows every time the team does, for a product that's fundamentally a database with a UI wrapped around it.

Worse, none of them were built with AI agents in mind as participants. Bots live behind an integration marketplace, boxed into slash commands and narrow webhooks — not a real presence in the room who can read a channel, understand context, and reply like anyone else on the team.

What We Built

So we built Pingo: one React/TypeScript codebase that ships everywhere the team already is, with AI agents wired in as first-class teammates from day one.

  • One codebase ships to Mac, iPhone, Android, and web — Electron wraps it for desktop, Capacitor for mobile — so there's one UI to build and one set of bugs to fix, not four apps drifting out of sync.
  • Realtime sync runs on Supabase Realtime, with row-level security enforcing channel membership at the database layer — a user can't query a channel they're not a member of, let alone read its messages.
  • Outgoing messages queue and retry instead of vanishing when the connection drops, and each channel catches up cleanly on reconnect, so a subway tunnel or flaky hotel wifi never costs a message.
  • Push notifications fire through a Postgres webhook that fans out to every registered device through an edge function — the database itself is the message bus, not a separate notification service to keep in sync.
  • A rate-limited, key-authenticated bot API lets AI agents post and reply in channels exactly like any other teammate — the founder's own AI assistant says good morning in #general every day.
  • A custom QA gallery mode renders hostile content on demand — script-injection attempts, RTL text, pathological formatting — so rendering bugs get caught in a test view, not in a live channel.

The Stack

React 19TypeScriptElectronCapacitorSupabase
Results

4 platforms

Mac, iPhone, Android, and web from one codebase

89 commits

Zero to a team-wide beta in about a month

1st-class AI

Agents post and reply through a bot API, not a bolted-on integration

$0 per seat

No recurring per-seat SaaS bill as the team grows

The database is the message bus.

Read the full build story

Want something like this for your business?

Tell us about your business and we'll show you exactly where a system like this fits — no pressure, no jargon.