Instant 1.0, a backend for AI-coded apps

Instant 1.0 has been released. The core message is to build the best backend for AI-coded apps. After 4 years of work, it launched as fully open source, centered on a multi-tenant database built on top of Postgres and a Clojure sync engine.

The product's first advantage is unlimited app creation. Instead of spinning up a VM when creating a project, it only adds a few database rows, so inactive apps incur no compute or memory cost, and the overhead is very small even when active. Thanks to this structure, apps don't freeze, and you can create an isolated backend instantly, on the order of a few hundred milliseconds. After creation, you get a public App ID and a private Admin Token.

The second advantage is the sync engine. Instant provides multiplayer, offline mode, and optimistic updates by default for every app. This gives you the real-time behavior and instant responsiveness required by modern apps like Linear, Notion, and Figma, without separate infrastructure. The example code is short too. On the frontend, you run relational queries with db.useQuery and send changes with db.transact, and it works even offline.

The third advantage is integration of additional services. auth, file storage, presence, and streams are provided by default, and files are treated as database rows rather than as separate S3 objects. This lets deletion rules be tied at the database level, reducing the multiple sources of truth problem that arises across multiple services.

The architecture can be summarized in three layers.

• Client SDK: resolves queries offline and reflects transactions instantly
• Clojure Backend: handles realtime sync, permissions, auth, presence, storage, and streams
• Multi-tenant Postgres: a single Postgres instance logically separated by App ID

The client-side storage uses IndexedDB, and the internal data model is a triple store. By storing data in the form [entity, attribute, value], relations and filters can be handled together, and Datalog is layered on top to evaluate queries client-side. Changes are not applied to the data immediately but are queued in a pending queue, so that changes can be rolled back even if the server returns a failure. For this, an immutable triple store and mutative are used.

Finally, Instant is programmatic. App creation, schema push, and permissions updates can be handled via the API or CLI, and the essay itself built its demo using the API. Developers can start a new project with npx create-instant-app, or attach an environment where an agent can immediately build features with npx skills add instantdb/skills.