Pure-Agent Apps

Not every agent-native app needs a full SaaS-style interface. Sometimes the agent is the product. A support triage agent. A daily report generator. A research bot. An email auto-responder. An ops runbook executor.

For these, the "app" is mostly just the agent doing work in the background. You still want a UI — but a minimal one, focused on observability, management, and steering rather than hand-crafted dashboards and forms.

This is the "agents benefit from a UI even when there's no rich app around them" pattern. The hot take is "agents will replace apps." The reality is "every agent eventually needs a UI for humans to supervise, configure, and debug it." Agent-native gives you that UI for free.

The minimum viable UI

A pure-agent app ships with five surfaces, all provided by the framework — you don't build them:

  1. Chat — the main input. Users talk to the agent, steer it, queue tasks. (<AgentSidebar> or <AgentPanel>)
  2. Workspace — skills, memory (learnings.md), AGENTS.md, custom sub-agents, connected MCP servers, scheduled jobs. Customize the agent's behavior without shipping code. (Workspace tab in the sidebar)
  3. Job history — which scheduled jobs ran, when, whether they succeeded, what they did. (Workspace tab → jobs/)
  4. Thread history — every past conversation, each preserved with its tool calls and final output. (Chat tab)
  5. Settings — API keys, connected accounts, onboarding status. (Sidebar settings)

Those five together are enough UI for most pure-agent use cases. No analytics dashboard. No Kanban. No forms. Just: talk to it, see what it's done, configure how it behaves.

When to pick this pattern

Pure-agent makes sense when:

  • The work happens in the background. Scheduled jobs, webhook-triggered handlers, Slack/Telegram responders. Users rarely sit in the app.
  • The output leaves the app. The agent posts to Slack, sends email, writes to a third-party system. There's nothing to view in-app; the value is elsewhere.
  • The domain is one-shot. A research bot that returns a report. No persistent object to dashboard.
  • You're prototyping. Ship the agent now; add a rich UI only when you've proven users need it.

Pick a full cloneable SaaS template instead when the app has real persistent objects (emails, events, documents, charts) users need to browse, pivot, and share.

The minimal scaffold

Start from the Starter template:

pnpm dlx @agent-native/core create my-agent --template starter

Starter gives you the six-rules architecture, the agent panel, the workspace, auth, polling, and one example action — and nothing else. Add your own actions in actions/, connect any MCP servers you need, write the relevant skills into the workspace, and you're done. The "UI" is the agent sidebar — which is already complete.

If you really want zero UI except the agent, app/routes/index.tsx can just render <AgentPanel defaultMode="chat" /> fullscreen. The only thing the user sees is the chat. Everything else — job history, workspace, settings — is one click away in the panel's tabs.

What you still get for free

Even with no custom UI, you still inherit every framework benefit:

  • Actions as agent tools + HTTP endpoints + MCP tools + A2A tools. External agents, Claude Desktop, and your own HTTP clients can drive the agent without going through the chat UI.
  • Recurring jobs for scheduled work — "every morning at 7 summarize my unread emails and post to Slack."
  • The workspace for per-user customization, skills, memory, MCP connections.
  • Sub-agent delegation via agent teams.
  • Portability — deploys to any serverless host, any SQL database.
  • Multi-tenant by default — each user gets their own workspace without a dev-box.

Adding a tiny bit of UI

Most "pure-agent" apps eventually want a little bit of custom UI — not a dashboard, but maybe a status page, a job history, or a config screen. The drop-in agent components coexist with anything else you render. Add a single /status route that lists recent runs; keep everything else in the chat. That's usually enough.

What's next

  • Recurring Jobs — scheduled prompts the agent runs on its own
  • Drop-in Agent — mounting <AgentPanel> fullscreen or in a sidebar
  • Actions — the tools your pure-agent will call
  • Workspace — the customization surface for skills, memory, and MCP servers