Skip to content

Agent AppExecutable intelligent applications.

A package standard for complete intelligent apps that run through shared host capabilities and App Server bridge contracts.

What Agent App defines

ContractQuestion answered
App packageWhat installable app is this and what does it contain?
EntriesWhich pages, commands, workflows, artifacts, background tasks, or settings can the host expose?
CapabilitiesWhich host standards and capability surfaces does this app need?
Bindings and contextWhich Context, Knowledge, Skill, Tool, Connector, Artifact, Evidence, Policy, and QC dependencies must the user or tenant satisfy?
ProjectionHow does a host compile the app into a catalog without inventing a second runtime?
ReadinessWhich runtime, connector, permission, evidence, and quality gates must pass before the app can run?
RoleStart hereThen read
App authorQuickstartRuntime package, manifest design, permissions, release.
Host implementorDesktop host conformanceRuntime model, Capability SDK, projection, readiness, security.
Standards reviewerSpecificationJSON Schemas, glossary, version notes.
Product plannerWhat is Agent App?Standards ecosystem boundary, examples, mini-program analogy.

Current Promise

A current app should be understandable before execution, installable without changing host core, runnable through typed capability handles, observable through RuntimeCore-derived Agent task events, explicit about delivery boundaries, clear about how users install it, and explicit about how lime.agent / lime.workflow calls enter App Server JSON-RPC. It should also follow the shared host model: user state and platform capabilities come from the host, while app-local storage and app backend services stay isolated. If a package cannot explain its entries, permissions, data boundary, runtime assets, task output contracts, external integrations, side effects, human approval boundaries, install mode, runtime bridge, and local storage placement, it is not ready to be distributed as an Agent App.

What a complete app page should answer

Every Agent App page in this documentation should help a reader answer four questions: what boundary this concept owns, which manifest or runtime fields express it, how a host should implement it, and what makes it ready for release. If a page only defines a term but does not show its implementation, readiness, or failure mode, treat it as incomplete.

Typical implementation sequence

  1. Define the app boundary and user-visible entries.
  2. Write the requirement, boundary, integration, and operation files.
  3. Write the install contract for in-host, standalone, runtime-backed, or web-host distribution.
  4. Write the App Server bridge profile for lime.agent / lime.workflow execution.
  5. Declare capabilities, storage, Runtime, UI, Context, Knowledge, Skills, Tools / Connectors, Artifacts, Evidence, Policy, QC, permissions, and Evals.
  6. Add runtime package assets and call host services through the Capability SDK.
  7. Build projection and readiness checks before enabling execution.
  8. Release with package hashes, compatibility metadata, overlays, install metadata, bridge metadata, and rollback guidance.

Draft host-platform standard for installable agent applications.