What is Agent App?

Agent App is a draft standard for complete installable applications in agent hosts. An app may include real UI, business workflows, data storage, background jobs, agent entries, Skills, Tools, Knowledge bindings, Artifacts, Policies, and Evals.

In one sentence: Agent App is an intelligent application running on Lime platform capabilities. It is not a Markdown file and not a single chat expert.

APP.md is only the discovery entry and manifest carrier. Real business capability comes from the runtime package and from calls through the Lime Capability SDK.

Relationship to experts

A large-company “expert” is usually:

Expert = Chat UI + Persona + Skills + Tools + Data Connections

Agent App is a higher-level package:

Agent App = UI + Workflow + Storage + Services + Agent Entries + Skills + Tools + Knowledge + Artifacts + Policy

An expert is only an expert-chat entry. An app can contain multiple experts, or none. An AI content engineering app should have project home, knowledge pages, content factory, review dashboard, and background jobs; a content strategist expert is only one entry.

Mini-program analogy

Agent App can be understood as a mini-program-like model for AI agents, without copying the WeChat Mini Program framework.

Mini-program mental model	Agent App counterpart
WeChat is the host platform.	Lime / IDE / AI Client is the host platform.
Mini-programs declare pages, components, permissions, storage.	Agent Apps declare UI, entries, capabilities, storage, permissions.
Mini-programs call wx.*.	Agent Apps call lime.ui, lime.storage, lime.agent, etc. through @lime/app-sdk.
The platform manages review, release, and permissions.	Cloud / Registry manages release, tenant enablement, license, policy.
The client runs the mini-program.	Lime Desktop installs and runs the app package locally.

The important part is not how it looks, but that the host opens capabilities and apps call those capabilities through a stable SDK.

Position in Lime

flowchart TD
  Cloud[Lime Cloud
Catalog / Release / License / Tenant Enablement] --> Desktop[Lime Desktop
Install / Cache / Resolver]
  Desktop --> Bridge[Capability Bridge]
  Bridge --> SDK[@lime/app-sdk]
  SDK --> App[Agent App Runtime Package
UI / Worker / Workflow / Storage]
  Bridge --> UI[Lime UI]
  Bridge --> Storage[Lime Storage]
  Bridge --> Runtime[Local Agent Runtime]
  Bridge --> Knowledge[Agent Knowledge]
  Bridge --> Tools[Tool Broker / ToolHub]
  Bridge --> Artifact[Agent Artifact]
  Bridge --> Evidence[Agent Evidence]
  Bridge --> Policy[Policy / Secrets]

Lime Cloud may distribute, authorize, and enable Agent Apps. Lime Desktop installs, authorizes, injects capabilities, and runs them locally. Cloud should not become a hidden Agent Runtime by default.

Good fits

• AI content engineering systems.
• Customer support knowledge workbenches.
• Sales SOP applications.
• Legal contract review products.
• Investment research workbenches.
• Internal enterprise workflow apps.
• Customer-specific private business systems.

These scenarios should not be implemented by changing Lime Core. New scenarios should become Agent Apps that call Lime platform capabilities.

Non-goals

Agent App is not:

• a collection of APP.md documents
• a single Expert or Persona
• a replacement for SKILL.md
• a knowledge base format
• a tool protocol
• a cloud Agent Runtime
• a package containing customer private data

Why it exists

Skills, Knowledge, and Tools are not enough for real business applications. Apps also need:

• their own UI pages, panels, and settings
• their own data models, indexes, migrations, and caches
• business workflows, background jobs, and human review nodes
• multiple chat or non-chat entries
• traceable Artifacts, Evidence, and Evals
• permissions, costs, credentials, tenant overlays, upgrade policies
• an SDK boundary that decouples apps from Lime internals

Agent App is the application layer that ties these pieces together.