ViteHub is still experimental. Expect bugs and breaking changes.

Overview

Learn the ViteHub concepts that connect server primitives, agents, integrations, and runtime inspection.

ViteHub is easiest to understand as a small set of boundaries. Server primitives give application code stable runtime APIs, while Agent Definitions compose Agent Drivers, Capabilities, Workspaces, Sources, Agent Invokers, and Agent Invocations.

The concepts in this section explain those boundaries before the package pages add options and examples. Read them when a feature crosses package ownership, host output, model-facing access, or runtime inspection.

Concept map

Concept	Use it to understand
Server primitives for any host	Why ViteHub starts from host-independent server behavior.
How ViteHub fits together	How Vite Integrations, Definitions, Provider Output, Runtime Helpers, and Capabilities connect.
Definitions and discovery	How package-owned files become named runtime behavior.
Vite Integrations and Provider Output	What build and dev integrations own.
Runtime Helpers and stable imports	Why application code imports stable ViteHub APIs instead of generated internals.
Workspace and Sources	How persistent file trees consume read-only origins.
Capabilities API	How Agents receive model-facing abilities.
Channels API	How message-shaped Agent Invocations, channel metadata, and host commands stay separate.
Auth Users and Agent Invokers	How authenticated app users map into trusted Agent Invocation identity.
Runtime policy, approvals, and traces	How ViteHub records runtime decisions without making policy invisible.

Reading order

Start with the first three pages when you are new to the project. Jump to Workspace, Capabilities, Auth, or runtime policy when you are adding those surfaces to an Agent Definition.

Next steps

Continue with Server primitives for any host.
Open Installation when you want a runnable setup.
Open Server primitives or Agents when you already know the boundary you need.