Build SaaS where the agent and UI share state

What it does

Agent-Native builds SaaS applications that are simultaneously rich web interfaces and autonomous agents. The UI and the agent share one SQL database and one state, so clicking a button and sending a prompt trigger the same underlying actions. It ships with a suite of cloneable, production-grade templates—Mail, Calendar, Brain, Analytics, and others—meant to replace paid tools you already use.

The interesting bit

Instead of bolting a chatbot onto a separate frontend, the architecture makes the agent and the UI roommates with a shared lease: one database, one state, and mutual awareness of what the other is doing. The agent knows what the user is looking at—select text and hit Cmd+I to prompt in context—and can even modify the application’s own features and UI over time.

Key highlights

• One shared SQL database and state between agent and UI; changes from either surface instantly on the other.
• Context-aware agents that know what the user is viewing, with Cmd+I prompts on selected text.
• Per-user workspaces for skills, memory, instructions, and MCP servers, stored in SQL and customizable per user.
• Cross-agent delegation via A2A, plus reusable provider connections and vault secrets managed through Dispatch.
• Runs on any SQL database Drizzle supports and any host Nitro supports; works with Claude Code, Codex, Gemini CLI, or Builder.io.

Caveats

• The CLI documentation references legacy commands (dev:all, --app-surface) and older “eager” modes alongside newer ones, suggesting the developer experience is still settling.

Verdict

Teams tired of renting SaaS should look here: you get cloneable, self-hosted Mail, Calendar, Analytics, and more that you can customize with an agent. Skip it if you want a lightweight library; this is an opinionated full-stack framework that ships with a dozen complete SaaS templates and its own coding agent.