@copilotkit/channels review
@copilotkit/channels 0.11.1 bundles the channel engine, JSX UI and five platform adapters. Its npm dependencies pin core, UI, Slack, Teams, Discord, Telegram and WhatsApp to 0.11.1. The supplied main-branch README says Channels run through a runner. Installing the bundle alone does not configure that runner, provider access or an agent. The examples below explain the supplied documentation; they were not executed or verified against the installed release.
Consider @copilotkit/channels 0.11.1 when the bundled dependency set is useful and your team can configure and verify the runner path. The release metadata proves the package pins; the supplied main-branch examples explain integration concepts. They do not establish release compatibility or successful recovery.
Answers from our run
@copilotkit/channels or @copilotkit/channels-core: which should you use?
@copilotkit/channels-core: Use core plus a direct adapter for adapter authoring or a selective dependency graph. The umbrella README documents that installation path; it does not establish a measured bundle-size saving. Consider @copilotkit/channels 0.11.1 when the bundled dependency set is useful and your team can configure and verify the runner path.
When should you not use @copilotkit/channels?
You need installation alone to produce a connected bot; the documented architecture also requires runner and provider configuration.
Use it if
- You want core, UI and five adapter dependencies pinned together in one package.
- You can configure provider access, application user identity and a channel runner.
- Your team can inspect the installed release declarations and test its delivery and persistence behavior before deployment.
- You need installation alone to produce a connected bot; the documented architecture also requires runner and provider configuration.
- You require verified compatibility with runtime 1.79.1; the packet does not establish that pairing.
- You require measured recovery, ordering or delivery guarantees; no matching lab measurement was supplied.
- You only need Slack listeners and prefer to connect the agent yourself; Bolt is an alternative.
- You want a selective dependency graph; core plus a direct adapter is an alternative.
Setup reality
@copilotkit/channels 0.11.1 bundles the channel engine, JSX UI and five platform adapters. The pinned npm record establishes its dependency versions and export paths. The integration READMEs and implementation extracts supplied for this guide come from main. They explain the documented architecture but do not establish that those APIs match the installed 0.11.1 release. No snippet here was executed, typechecked or connected to a provider, and no matching lab measurement was supplied.
The supplied main-branch README says Channels run through a runner. It documents CopilotKit Intelligence as the managed path and also describes operating a custom runner. On the custom path, your team owns state, persistence, concurrency, locking, retries and race handling. Provider access and agent configuration remain separate work. The README's runtime example is therefore architectural guidance, not a verified application for Channels 0.11.1 and runtime 1.79.1. Check matching release declarations before assembling that integration.
In the supplied main implementation, a message-capable adapter requires an onMention or onMessage handler before activation. The channel declaration below is only the first step. Pair it with one of the message-handler examples before a runner activates a message-capable adapter. The history tool and thread-start handler do not satisfy that requirement by themselves. This distinction matters when copying a small example into a larger application: registering a tool does not register an incoming-message handler.
In the supplied main implementation, an explicit store.adapter wins; otherwise it uses the first adapter-provided stateStore, then MemoryStore. If several adapters supply a store, the implementation warns and chooses the first. Choose store.adapter explicitly when that selection would be ambiguous. MemoryStore loses state on restart. The core README's actionStore instructions describe a different persistence contract from the supplied implementation, where actionStore is deprecated in favor of store.adapter. Reconcile that difference with the installed release before configuring persistence.
The source describes parallel, serial and drop concurrency modes, but the extracts do not demonstrate the complete execution path or backend behavior. This guide provides no read-modify-write counter or claim that a concurrency setting makes such a counter safe. Verify overlapping deliveries, restart behavior and the actual provider path in your own trial.
Ratings use a source-review rubric: 1 means adoption-blocking evidence gaps, 2 substantial reconciliation work, 3 useful documented support with material limits, 4 strong evidence with small gaps, and 5 support demonstrated across releases and deployments. The numbers express confidence supported by this packet; they do not measure uptime, adoption or operational quality. In particular, the maintenance score is provisional because one pinned release cannot establish a maintenance history. Start with one provider, inspect the installed contracts, and test a restart and overlapping messages before widening the deployment.
Patterns
Configure the documented JSX runtime configure-channel-jsx
{
"compilerOptions": {
"jsx": "react-jsx",
"jsxImportSource": "@copilotkit/channels"
}
}Unexecuted configuration from the supplied main-branch umbrella README. Apply it to the TypeScript configuration compiling channel .tsx files. It is not a complete project configuration verified against 0.11.1.
Declare the channel before adding a handler declare-channel
import { createChannel } from '@copilotkit/channels'
export const channel = createChannel({
name: 'support-bot',
identifyUser: 'platform'
})Unexecuted illustration of the supplied main API. This declaration has no adapter, runner or agent. Add an onMessage or onMention handler before activating a message-capable adapter; the supplied implementation rejects activation without one. Check the installed release declarations before use.
Register an incoming-message handler register-message-handler
import { channel } from './channel.js'
channel.onMessage(async ({ thread, message }) => {
await thread.post(message.text)
})Unexecuted illustration adapted from the main documentation. Register it before activation. This handler supplies the incoming-message step missing from the bare declaration, but it does not connect a provider. Choose one onMessage example for this illustration rather than registering all the examples together.
Post a JSX reply render-message-jsx
import { Message, Section } from '@copilotkit/channels'
import { channel } from './channel.js'
channel.onMessage(async ({ thread, message }) => {
await thread.post(
<Message>
<Section>Echo: {message.text}</Section>
</Message>
)
})Unexecuted illustration adapted from the main-branch umbrella README. Use a .tsx file with the Channels JSX setting. This replaces the plain-text echo handler in the preceding example. Equal rendering across providers was not tested.
Pass the posted reference to update update-posted-message
import { channel } from './channel.js'
channel.onMessage(async ({ thread }) => {
const ref = await thread.post('Preparing a reply...')
await thread.update(ref, 'Reply prepared.')
})Unexecuted illustration of the post and update signatures in the main-branch core README. Keep the MessageRef returned by post and pass it to update. Provider update behavior was not tested. This is another alternative onMessage handler.
Stream fixed example text stream-text-reply
import { channel } from './channel.js'
async function* chunks() {
yield 'An example '
yield 'reply.'
}
channel.onMessage(async ({ thread }) => {
await thread.stream(chunks())
})Unexecuted illustration of the main-branch Thread interface, which accepts AsyncIterable<string>. The generator supplies fixed text and makes no model request. It demonstrates the documented input shape without establishing delivery latency or rendering behavior.
Register a history-reading tool register-thread-tool
import { defineChannelTool } from '@copilotkit/channels'
import { z } from 'zod'
import { channel } from './channel.js'
channel.tool(defineChannelTool({
name: 'read_current_thread',
description: 'Read messages available in the current conversation.',
parameters: z.object({}),
async handler(_args, { thread }) {
return await thread.getMessages()
}
}))Unexecuted illustration adapted from the main-branch core README. This form requires Zod as a separate application dependency. Register the tool before activation and configure an agent to call it. History depends on provider capability; the packet does not establish the unsupported-operation return value. This tool alone does not supply the message handler required for activation with a message-capable adapter.
Handle an unsupported title operation handle-optional-capability
import { channel } from './channel.js'
channel.onThreadStarted(async ({ thread }) => {
const result = await thread.setTitle('Support conversation')
if (!result.ok) {
await thread.post('Support conversation')
}
})Unexecuted illustration based on the main-branch core README. Its Thread interface reports ok: false on surfaces without title support. Adapters without the thread-start concept never invoke this handler, so the fallback only runs after that event occurs. Add an ordinary message handler separately when activating a message-capable adapter.
Alternatives
| Package | Registry | Pick it when |
|---|---|---|
| @copilotkit/channels-core | npm | Use core plus a direct adapter for adapter authoring or a selective dependency graph. The umbrella README documents that installation path; it does not establish a measured bundle-size saving. |
| @slack/bolt | npm | Bolt documents listeners for Slack events, actions and slash commands. Consider it when those Slack-specific handlers fit your application and you can supply the agent integration. No performance or setup-time comparison was measured. |
More ai / ml guides
openai · mcp · huggingface-hub · scikit-learn · tiktoken · tokenizers · the whole shelf →
How this guide is made: grounded in the library's documentation, release notes, changelog, and issue history, on a fixed rubric — not a hands-on install of every release. The 50 most-downloaded entries are additionally install-verified in clean containers. Corrections: contact the desk.