> ## Documentation Index
> Fetch the complete documentation index at: https://docs.cogniagent.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Channels

> Where people talk to an agent: its phone numbers, website chat, inboxes and team chats, each one answering people or starting conversations.

An agent's channels are where people talk to it: a phone number, WhatsApp, the chat on your website, an inbox, Slack, Teams, Telegram or SMS. Open the agent, click **Settings** in its header, and choose **Channels**.

What other agents, your apps and AI assistants may call is a separate section, **External access**. See [Endpoints](https://docs.cogniagent.ai/features/endpoints.md). What starts the agent's work on its own, such as webhooks and schedules, is in **Triggers**. See [Triggers](https://docs.cogniagent.ai/features/agents.md#triggers).

**Unsorted** has Channels too, for the conversation flows and coworkers in it. It has no External access: every agent can already use what is in Unsorted.

## The list

Every address people reach the agent on is a row. A row shows the channel's logo and its address, **Answers** or **Starts**, who talks on it, and an on/off switch. Click a row to open its drawer.

A row's switch leaves one channel out. The agent's own **Start** and **Stop**, in its header, open and close every channel at once (see [Start and Stop](https://docs.cogniagent.ai/features/agents.md#start-and-stop)):

* **While the agent runs,** a row you add or switch on answers at once, and a row you switch off stops at once.
* **While it is stopped,** no channel answers or starts, and a row that is on says **Waits for the agent to start**. Saving a row says *Saved. It goes live when the agent starts.*

### Answers and starts

* **Answers.** Whoever writes or calls is answered.
* **Starts.** The agent opens conversations from this address, and you give the recipient each time.

One conversation flow can answer on one row and start from another. Choosing **Both** when you add a channel makes two rows: one answers, one starts.

**Who talks** is one of this agent's conversation flows or coworkers. A coworker on a **Starts** row starts conversations through **Talk to people**, and each start asks a person first. Choices that can't talk on that kind of channel yet are greyed **Later**.

An address belongs to one agent, and in each environment only one row answers on it. If another row already answers there, you are told which one, with **Move it here** when you may change that row's agent. Copies of one channel in other environments share its address and need no **Move it here**: see [In each environment](#in-each-environment).

## Add a channel

Click **Add a channel**. A drawer opens at the side of the page, the same drawer you edit a channel in.

**1. The kind**

Pick the kind of channel, in three groups. Once picked, the grid folds to that kind, with **Change** to open it again.

| Group | Kinds |
| - | - |
| **Calls and messages** | **Phone** (A number people call), **WhatsApp** (Your workspace's business number), **Website chat** (On your site, from a code snippet), **SMS** (Text messages through Twilio) |
| **Email** | **Gmail** (A Google inbox it reads and answers), **Outlook** (A Microsoft inbox it reads and answers) |
| **Team chat** | **Slackbot** (Ready to use, nothing to set up), **Slack as a person** (Writes as your own Slack account), **Slack bot** (Your own app, with its token), **Teams** (Microsoft chats and channels), **Telegram** (Your bot, through its token) |

**2. The address**

Pick a number or a connection. **Add a number** opens Phone Numbers, and **Connect** connects a new account. A website chat is its own address.

You can also pick **From variable** to take the account from a Connection variable of the agent. The channel listens with the account the variable holds in the agent's current environment. If the variable has no usable account, the channel stays off and Start lists it. See [Connection variables](https://docs.cogniagent.ai/features/agent-variables.md#use-it-in-a-trigger-or-a-channel).

**3. Answers, starts, or both**

**Answers** (Replies to whoever writes), **Starts** (Opens conversations) or **Both** (Two channels: one answers, one starts). A choice the kind can't do is greyed, like "SMS cannot answer".

**4. Who talks**

Pick a conversation flow or coworker of this agent, then click **Add channel**.

The drawer then shows the new channel, ready for its settings. A channel a coworker answers on opens at **Who may write**, since nobody may write to it until you list someone.

Each kind of channel also has settings of its own in the drawer, such as the widget's look or the Slack channel to listen in. See the page for [each kind of channel](https://docs.cogniagent.ai/conversation-flows/channels/overview.md).

### From the flow editor

In the conversation flow editor, **Channels** (in the rail, or the **Channels** tab of **Agent settings**) lists the rows the flow talks on in the environment the agent runs in, under **Answers on** and **Starts from**. Click one to open its drawer without leaving the editor. **Add a channel** opens the same steps, with **Who talks** already set to this flow.

Channels still belong to the agent. Whatever you add or change in the flow editor shows in the agent's Channels too, and **Open Channels** takes you there.

## Channel settings

A row's drawer shows its address and who talks, then the settings below. **Save** keeps every change in the drawer at once. Leaving a drawer with changes you haven't saved asks first: **Discard unsaved changes?**, with **Discard** or **Keep editing**.

> [accordion] Start values
>
> Values a new conversation on this channel starts with. Each value is one of:
>
> * **Fixed**: the value you type. A fixed value wins over any other.
> * **Asked at start**: asked each time someone starts a conversation from this row. Only on a flow's **Starts** rows.
> * **From the channel**: read from the message, such as the sender's number, email or name, the Slack member or channel, or the Telegram user.
>
> A fixed value is not filled in from variables, and the box says so: "Sent exactly as typed." A short value, such as a number, an email or a date, takes one line. Longer text, such as a message, has room for several lines and opens full screen. A yes-or-no value is picked from a list.
>
> On a coworker's row: "Values a new session from this channel starts with, added to its first message."

> [accordion] Guardrails
>
> An optional extra layer, off until you add it. Each actor's own guardrails and the safety check always run; a coworker always follows its own instructions. Add rules here only for what this channel needs on top.
>
> * **What it declines**: topics it declines and things it never says. A fast check reads each incoming message, and each reply before it is sent, against this text.
> * **Decline reply (optional)**: what a declined message gets. After three declines in a conversation within an hour, it stays silent.
> * **Longest message** and **Messages per conversation per hour**: plain limits, checked without the AI.
>
> On the website chat and the phone, replies are spoken or streamed as they are made, so only incoming messages are checked. If the check doesn't answer in time, the message goes through, and the drawer counts how often that happened today.

> [accordion] Who may write
>
> On a row that answers: **Anyone**, or **Only these**:
>
> * **People**: their address on that channel, such as an email address, a phone number or a Slack member ID.
> * **Email domains**, on email rows.
> * **Slack channels**, on a conversation flow's Slack rows: anyone writing in them may write.
>
> Messages from anyone else are dropped before a conversation starts. Senders are matched on their address as written; it is not verified. The website chat and phone lines answer anyone.
>
> A row a coworker answers on starts with nobody listed, and says "Nobody may write yet." until you add someone or choose **Anyone**.

> [accordion] Capabilities on this channel
>
> On a coworker's row, every capability is on. Switch any of them off for this channel:
>
> * **Tools**: Web search, Read web pages, Generate images, Live browser, Code and files.
> * **What it can start**: Schedule itself, Talk to people, Collaboration, The agent's own tools.
>
> A switch-off holds on every turn of the sessions this channel starts, including wake-ups and messages you type into them. Work the coworker hands to another coworker keeps the same switch-offs.

## Coworkers on channels

A coworker's channels are set here, on its agent. The coworker's own page has no Channels tab.

* **Answers**: a coworker answers on Slack, Gmail and Telegram. What people write there reaches it as sessions you can follow.
* **Starts**: a coworker starts conversations on Slack and Gmail, through **Talk to people**. It is always on for every coworker, and these rows are the channels it may start from. Each start shows a person the recipient and the first message, and waits for their approval.

See [Message channels](https://docs.cogniagent.ai/cowork/capabilities/message-channels.md) and [Talking to people](https://docs.cogniagent.ai/cowork/capabilities/talking-to-people.md).

## Start a conversation

When one of the agent's conversation flows has a **Starts** row that can start now, the agent's page shows **Start a conversation**. The menu of each such row has it too.

1. **From**: pick the starting row.
2. **To**: type the recipient, in the channel's format. It is checked as you type.
3. Fill in the row's values asked at start. Fixed values show, locked.
4. Write the first message where the channel needs one: a subject and body for email, or the approved template on WhatsApp when the person hasn't written in the last 24 hours. A phone row calls right away.
5. Click **Start**. For a website chat, click **Create link** instead.

The conversation opens in a side panel. The row's drawer lists the conversations **Started from here**.

Conversations also start from a workflow's [Conversation flow](https://docs.cogniagent.ai/nodes/communication/conversation-flow.md) step, from an [endpoint](https://docs.cogniagent.ai/features/endpoints.md) whose target is a starting row, and from a coworker. A coworker always asks a person first.

## In each environment

When the agent has more than one [environment](https://docs.cogniagent.ai/features/environments.md), such as Production and Demo, each environment has its own channels. A channel belongs to one environment, with its own account, address, recipients and settings. Only the channels of the environment the agent runs in answer and start.

* **Channels shows one environment at a time.** Pick it under **Environment**: the one the agent runs in comes first, marked "(running now)". When it shows another one, a note says: "Demo's channels. They answer while the agent runs in Demo. It runs in Production now." You can change them there too.
* **Add a channel** adds it to the environment shown: "Adds it to Demo. It answers while the agent runs in Demo."
* **The drawer names its environment**, such as "Demo's channel. It answers while the agent runs in Demo, and is off now: the agent runs in Production."
* **The agent's home** lists the channels of the environment it runs in, and counts the others, such as "2 more in other environments". A problem with another environment's channel still shows, named with its environment, so you can fix it before you switch.

Before a [switch](https://docs.cogniagent.ai/features/environments.md#switch-the-environment), the confirm lists the channels that stop and start, such as "Stops answering on +1 555 0100 (Production's)" and "Starts answering on +1 555 0199 (Demo's)".

### Copies

[Adding an environment](https://docs.cogniagent.ai/features/environments.md#add-an-environment) can copy another environment's channels, together with its endpoints, trigger settings and schedules, or start with none. To copy one channel later, open its row's menu and click **Copy to Demo**. It is offered for each environment that has no copy of it yet.

A copy is a separate channel: changing it never changes the original.

* **A copy starts on the same address as its original.** Only the copy in the environment the agent runs in answers. The other one is off there, and its row says **Off in Demo**.
* **To use a test number, mailbox or Slack channel in Demo**, change Demo's copy. Production's keeps its own.
* **A website chat copy keeps its original's embed.** The same snippet shows Demo's widget while the agent runs in Demo. For a separate test widget, add a new website chat channel in Demo.

### Workflows, endpoints and Human Steps

When a workflow's [Conversation flow](https://docs.cogniagent.ai/nodes/communication/conversation-flow.md) step, an [endpoint](https://docs.cogniagent.ai/features/endpoints.md) or a [Human Step](https://docs.cogniagent.ai/nodes/communication/human-step.md) uses a channel, it starts on that channel's copy in the environment the agent runs in. A step saved on any copy works in every environment that has one.

If that environment has no copy, the start is refused with `channel_not_in_environment`: "Support line has no channel in Demo. Copy it to Demo on the Channels tab." The workflow step shows it as "No copy in Demo: Gmail" until you copy the channel or pick another one.

## Moving and deleting

* **Moving** a conversation flow or coworker to another agent asks about its rows: they move with it, or stay here, switched off, with no one answering. Rows that move follow their new agent: they answer while it runs, and wait while it is stopped.
* **Only Production's rows move**, into the new agent's Production. The item's rows in other environments stay here, off, with no one answering. The Move dialog names them: 'Only Production's channels move. "Support line" (Demo) stays in Front desk, off, with no one answering.'
* **Deleting** a conversation flow or coworker keeps its rows in Channels, switched off, marked **No one answers**. The delete dialog lists them.
* **Deleting an environment** deletes its channels, with its endpoints, trigger settings and schedules. Their copies in other environments stay.

## Channels and workflows

A channel is for conversations: a flow or coworker answers on it, or starts from it, and keeps talking. A workflow reaches people with its own steps, through your [connections](https://docs.cogniagent.ai/features/integrations.md), and needs no channel here:

| To | Use |
| - | - |
| Start a workflow when a message arrives | [App Trigger](https://docs.cogniagent.ai/nodes/triggers/app-trigger.md), such as a new email or a Slack message |
| Send one message | [Integration Action](https://docs.cogniagent.ai/nodes/actions/integration-action.md), such as sending an email or a Slack message |
| Ask a person and wait for the answer | [Human Step](https://docs.cogniagent.ai/nodes/communication/human-step.md) |
| Hold a conversation | The [Conversation flow](https://docs.cogniagent.ai/nodes/communication/conversation-flow.md) step, which starts one from a **Starts** channel here |

## Who can change it

People who may change the agent see **Add a channel** and the switches, here and in the flow editor. Everyone else who can open the agent sees the rows, each marked **On** or **Off**.

## Common questions

> [accordion] Can I add a channel without leaving the flow editor?
>
> Yes. The flow editor's **Channels** has **Add a channel**, with **Who talks** set to the flow, and opens each row's drawer in place. The channels are the agent's rows, so its Channels settings list them too.

> [accordion] Where did a coworker's Channels tab go?
>
> A coworker's channels are rows in its agent's Channels, like a flow's. Add one with **Add a channel** and pick the coworker in **Who talks**.

> [accordion] Can a coworker answer the phone?
>
> No. A coworker answers on Slack, Gmail and Telegram, and starts conversations on Slack and Gmail. On other kinds of channel it is greyed **Later** in **Who talks**. A conversation flow can answer the phone.

> [accordion] Do guardrails cost anything?
>
> Only when you set them. A row with no guardrail text and no limits adds no check at all.

> [accordion] Why is a channel I set up in the flow editor missing here?
>
> Channels of the agent's flows that are not rows yet keep working as they are. A notice above the rows counts them, and someone who may change the agent can bring them in with one click.

> [accordion] Where are the webhooks?
>
> Not in Channels. A webhook is a trigger: the agent's **Settings → [Triggers](https://docs.cogniagent.ai/features/agents.md#triggers)** lists each one, where you can change it, add a token or switch it off. It is made in its workflow's [Webhook](https://docs.cogniagent.ai/nodes/triggers/webhook.md) step.

## Next steps

> [card] [Endpoints](https://docs.cogniagent.ai/features/endpoints.md)
>
> What the agent offers other agents, apps and AI assistants.

> [card] [Each kind of channel](https://docs.cogniagent.ai/conversation-flows/channels/overview.md)
>
> Each kind of channel and its own settings.

> [card] [Message channels](https://docs.cogniagent.ai/cowork/capabilities/message-channels.md)
>
> People reaching a coworker on Slack, Gmail or Telegram.

> [card] [Agents](https://docs.cogniagent.ai/features/agents.md)
>
> What an agent holds, and moving items between agents.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.