Learn how to give AI agents a stateful bash terminal and virtual file system using Convex, without requiring containers or VMs.

Hamza, shares a year-long build from 10 tables to 70+ Convex modules serving web, mobile, and REST APIs through a single deployment with one backend handling everything.

The video walks through an experiment in “de-lock-in-ifying” a small Convex app: starting with the basic TanStack Start template, the author recreates Convex queries, mutations and actions as TanStack Start server functions; swaps Convex’s reactive data layer for React Query (with manual cache invalidation); and replaces Convex’s built-in cloud database with a self-hosted Postgres instance accessed via Drizzle ORM—eventually wrapping Drizzle in a Convex-style API so most original code can be copy-pasted. They also bolt on transactions, discuss substitutes for other Convex features (file storage, realtime, auth, scheduling, search, etc.), and note that exporting Convex data is straightforward. The upshot: you can migrate off Convex without huge code changes, but you trade Convex’s “batteries-included” simplicity for extra infrastructure to manage—so the easiest escape hatch is still running Convex in self-hosted mode.

Building an AI chat app? Convex’s Persistent Text Streaming keeps conversations flowing—even across page reloads or tab switches—by storing and syncing chat data in real time. This video shows how to stream responses live, reduce bandwidth waste, and boost performance using Convex’s serverless backend. Say goodbye to broken chat experiences and hello to seamless, reliable interactions.