One backend container your AI coding agent actually drives

What it does Nubase packs PostgreSQL, authentication, object storage, edge functions, static asset hosting, cron, an AI gateway, and a Mem0-style memory service into a single self-hostable Java backend. It exposes a PostgREST-compatible REST API and a web Studio, though the intended interface is AI coding agents like Claude Code and Codex, which manage projects through an MCP bridge rather than clicking through dashboards.

The interesting bit Where Supabase’s open-source stack centers on one project per instance, Nubase treats multi-tenancy as the default: a single control plane spins up isolated project databases, each with its own auth, storage, and memory namespaces. That architectural shift matters because it lets an agent generate a full-stack app, deploy the frontend to Nubase’s built-in CDN, schedule cron jobs, and write durable user memory—all within the same conversation, on your own hardware.

Key highlights

• First-class Memory primitive with LLM-powered fact extraction and hybrid retrieval (pgvector + Postgres full-text + entity boost), not a bolt-on vector store.
• MCP tooling: agents inspect schema, run SQL, deploy functions, upload assets, and manage cron through a Model Context Protocol bridge.
• Multi-project self-hosting from one Docker container; per-project Postgres isolation with separate JWT secrets and row-level security.
• PostgREST-compatible REST API, Supabase-style auth (MFA, OAuth, SAML), and S3-compatible storage.
• Built-in Assets CDN and Functions runtime (local or Cloudflare Workers for Platforms) so generated apps can go live without external hosting accounts.

Caveats

• Realtime WebSocket subscriptions are listed as “Not yet” in the project’s own comparison table.
• First-boot secrets are auto-generated into a Docker volume; the README warns you must pin stable secrets manually for production deployments that survive restarts.
• It is a substantial Java/Spring Boot monolith bundling Redis and Postgres—expect resource overhead, not a lightweight single-binary edge runtime.

Verdict Worth evaluating if you self-host AI agent workflows and want one platform for auth, memory, storage, and deployment instead of a dozen separate services. Skip it if you need mature realtime push or a fully managed cloud SLA today.