ViteHub
ViteHub is still experimental. Expect bugs and breaking changes.

Channels API

Understand message-shaped Agent Invocations, channel metadata, Chat Capability behavior, and host-owned commands.

The Channels API names message-shaped reachability without turning delivery metadata into trusted identity. It keeps channel facts, message input, Chat Capability behavior, and host-owned commands separate.

Message-shaped input is one way to start an Agent Invocation. ViteHub keeps it separate from the Agent itself: an Agent can receive plain prompts, structured input, or messages depending on the trigger and Capability behavior.

The broader Channel language is still narrow in the current package surface. Today, channelId is Agent Run metadata supplied by a host or adapter, while the Chat Capability owns the chat.message trigger and Chat History behavior.

Chat Capability

The Chat Capability gives an Agent chat-oriented runtime behavior. It can contribute the chat.message Agent Trigger, Chat History handling, Chat Session selection, Chat Platform Adapter webhooks, and chat-derived Agent Invoker defaults.

Application routes can consume the same trigger directly when the app owns the chat UI.

server/api/support-chat.post.ts
import { runAgentTrigger } from '@vite-hub/agent'
import support from '../agents/support'

export default defineEventHandler(async (event) => {
  const body = await readBody<{ text: string }>(event)

  return runAgentTrigger(support, { runtime: 'vite' }, 'chat.message', {
    messages: [{
      id: crypto.randomUUID(),
      parts: [{ text: body.text, type: 'text' }],
      role: 'user',
    }],
  })
})

Message concepts

ConceptMeaning
ChannelHost-provided origin and delivery context, such as an app chat, GitHub thread, or external platform conversation.
Agent InvocationOne runtime request to an Agent. It may or may not be message-shaped.
MessageA normalized chat-style record with a role and parts such as text, tool calls, approvals, data, audio, sources, or errors.
Chat HistoryOrdered conversational messages for one chat interaction.
Chat History WindowThe bounded prior messages included in one Agent Invocation.
Chat SessionA host-visible conversation boundary inside Chat History.
Agent Run metadataHost-provided metadata such as origin, run id, thread id, or channel id.

Commands are not all the same

Input Commands are Capability behavior that transforms or enriches explicit user input before the Agent runs. Host Commands change chat, session, UI, or product state around an Agent and stay outside the Capability Lifecycle.

For example, a summary command can be an Input Command when it produces Agent run input. A /clear chat action is a Host Command because it mutates host chat state.

Inspect it

Inspect the trigger id, normalized messages, and Agent Run metadata. DevTools can show message-shaped runs for chat-capable Agents, while server routes can log the exact trigger input they pass to runAgentTrigger().

Next steps

Copyright © 2026