> AI agents: this is one page from PostHog's docs. Full index of Markdown docs for LLMs: https://posthog.com/llms.txt

# AI agent

Canonical URL: https://posthog.com/docs/support/ai-agent

## On this page

- [Turn it on](#turn-it-on)
- [Let it investigate the ticket](#let-it-investigate-the-ticket)
- [Choose the channels](#choose-the-channels)
- [Choose what the customer sees](#choose-what-the-customer-sees)

**Private beta**

The AI agent is in beta, and it is not enabled for every project. If you do not see **AI agent** under **Support** > **Settings**, ask us to turn it on. It is free while it is in beta. We will set a price later, and we will tell you before any charge starts.

The AI agent drafts a reply when a new ticket arrives. You decide whether that draft stays a private note for your team, or goes to the customer. It does not reply on its own until you choose **AI reply** for that channel and ticket type.

Replies are grounded in your [business knowledge](https://posthog.com/docs/business-knowledge.md) sources: documents, links, and other context you add.

![AI agent](https://res.cloudinary.com/dmukukwp6/image/upload/w_1600,c_limit,q_auto,f_auto/Screenshot_2026_10_20_at_17_32_11_4ca06d4936.png)

## Turn it on

1.  Go to **Support** > **Settings** > **AI agent**.
2.  Turn on **Enable AI agent**.
3.  Confirm AI data processing is allowed for your organization.
4.  Add at least one business knowledge source and wait until it is ready.

![Turn it on](https://res.cloudinary.com/dmukukwp6/image/upload/w_1600,c_limit,q_auto,f_auto/Screenshot_2026_10_20_at_17_28_16_52dfe7e34c.png)

Without consent and a ready source, the agent has nothing to ground a reply in.

## Let it investigate the ticket

Under **Investigate ticket data**, the **Allow the agent to investigate ticket data** switch is separate, and it starts off.

When it is on, a ticket that reports something broken can be checked against your project's events, [error tracking](https://posthog.com/docs/error-tracking.md) issues, [session replays](https://posthog.com/docs/session-replay.md), and [logs](https://posthog.com/docs/logs.md). The agent has read-only access, and only to that project.

Leave it off if you want replies grounded only in documentation and business knowledge.

![Let it investigate the ticket](https://res.cloudinary.com/dmukukwp6/image/upload/w_1600,c_limit,q_auto,f_auto/Screenshot_2026_10_20_at_17_28_20_ed671ea867.png)

## Choose the channels

Under **Allowed channels**, pick where the agent runs. A channel that is not connected yet stays disabled until you enable it under **Channels**. The list is the widget (including the API), Slack, email, Microsoft Teams, and GitHub.

![Choose the channels](https://res.cloudinary.com/dmukukwp6/image/upload/w_1600,c_limit,q_auto,f_auto/Screenshot_2026_10_20_at_17_28_42_e502308ad9.png)

## Choose what the customer sees

For each channel you enabled, set a reply mode per ticket type.

| Ticket type | What you can choose | Why |
| --- | --- | --- |
| **How-to** | **Private note** or **AI reply** | The customer needs instructions. A reviewed reply can go out. |
| **Diagnostic** | Private note only | The draft can include data from your project. |
| **Account/Billing** | Private note only | The draft can include account data. |

**Private note** is the default. Only your team sees it. **AI reply** sends a safety-reviewed message to the customer. On a channel set to **AI reply**, the agent may also ask the customer one clarifying question.

![Choose what the customer sees](https://res.cloudinary.com/dmukukwp6/image/upload/w_1600,c_limit,q_auto,f_auto/Screenshot_2026_10_20_at_17_28_49_ba946a7d3b.png)

Diagnostic and Account/Billing stay private even if you would prefer a customer reply. The product enforces that. Do not paste a private note into the customer thread if it contains project data you would not send yourself.