# Inbox agent

> When on, the AY Inbox inbox agent drafts a reply in your voice to each human reply your campaigns get, for you to approve, edit or discard. It never sends.

Source: https://docs.ayinbox.com/guides/inbox-agent/

The inbox agent watches your replies. When a real person replies to one of your campaigns, it reads the thread, drafts a suggested reply in your [voice](/guides/unibox/#ai-drafts), and attaches it to the conversation for you to review. You approve and send it, edit it first, or discard it. Nothing is ever sent because the agent wrote it: a person decides every time. A campaign can separately opt into [auto-reply](/guides/auto-reply/), which takes that campaign's qualifying replies off the draft path and answers them itself.

This is a paid feature and it is off until you turn it on.

Turning it on [#turning-it-on]

Open **Settings, Workspace** and find **Inbox agent** (it needs the **Manage settings** permission). Toggle **Draft replies for me** on. From then on, every inbound human reply to a campaign gets a draft.

<Callout type="info" title="Paid and opt-in">
  The inbox agent is available on paid plans and is off by default. Turning it on is an explicit choice, and each reply it drafts costs 5 AI credits. See [AI credits](/guides/ai-credits/).
</Callout>

What triggers a draft [#what-triggers-a-draft]

A draft is created when all of these are true:

* A person replies to one of your campaign emails (automated replies like out-of-office and bounces never trigger it).
* Your workspace is on a paid plan and has the inbox agent turned on.
* You have at least 5 AI credits.
* The reply has enough substance to answer. A one-line acknowledgement ("thanks", "got it") is skipped, so you are never charged to draft a reply to a throwaway line.

The agent does not start new conversations and never messages a new recipient. It only ever drafts a reply to someone who already replied to you.

Reviewing a draft [#reviewing-a-draft]

A new draft appears in three places:

* In the conversation itself: open the thread in the unibox and the draft appears below the messages, in an editable box, with **Approve and send**, and **Discard**.
* In the scope rail: the **Agent drafts** filter lists every conversation with a draft waiting, with a live count next to it. The count and list update the moment a new draft lands, no refresh needed.
* In the **Drafts** folder: the same conversations are listed first, under **Agent drafts**, above the drafts saved in your mailboxes, and the Drafts count includes them.

From the draft box you can:

* **Edit** the text right there. The box is editable, so tweak a line, fix a detail, or rewrite it before sending.
* **Approve and send.** Sends the text as it stands (with your edits) through the normal reply path, from the same mailbox, threaded to their reply. A reply in a [shared reply inbox](/guides/mailboxes/#a-shared-reply-inbox) is answered from the mailbox that emailed them instead. This is the only action that sends.
* **Discard.** Dismisses the draft without sending. Use it when the reply is better handled by hand.

Each thread holds at most one pending draft at a time, and once you approve or discard it the draft is resolved.

What the draft is grounded in [#what-the-draft-is-grounded-in]

The agent writes with the same context as an [AI reply draft](/guides/unibox/#ai-drafts): the full thread so far, read as the messages themselves rather than their preview lines, what AY Inbox knows about the contact (their name, company, and any custom fields on their record), the campaign and the step the prospect answered, and the [campaign brain](/guides/campaign-brain/) that applies (what you sell, pricing, links, rules and material). With no brain it reads your workspace [voice profile](/guides/unibox/#ai-drafts) and lists your [enabled skills](/guides/ai-skills/); with a brain, the skills that match the reply are read with it.

Under the draft, the **Used** line shows the campaign, the brain and each source the draft read, and badges above **Approve and send** show anything the [check](/guides/campaign-brain/#flags) flagged: a link, price or phrase the brain does not give. Flags never stop you sending. The agent still never sends on its own.

When you change a draft before approving it, AY Inbox records that it was edited. [Auto-reply](/guides/auto-reply/#modes) reads that to show how often Supervised answers went out as written.

Credits and failures [#credits-and-failures]

* **5 credits per handled reply**, charged when a draft is created. If the model is unavailable, no draft is created and nothing is charged.
* **Out of credits.** When your balance is below 5, the agent simply does not draft that reply. Your inbox and campaigns are unaffected; you just handle that one by hand. Top up under **Settings, Billing**.
* **The agent never sends.** There is no setting that lets the agent send for you; approving its draft is always a person's explicit action. Answering automatically is a separate, per-campaign opt-in: [auto-reply](/guides/auto-reply/).

From an agent [#from-an-agent]

An [MCP client](/api/mcp/) or [agent tool](/api/agent-tools/) caller reads the waiting drafts with `list_agent_drafts` (`READ_UNIBOX`) and closes one unsent with `discard_agent_draft` (`WRITE_UNIBOX`). Approving one sends it, so that stays with a person, in the dashboard or `POST /unibox/agent-drafts/:id/approve`. An agent that wants its own draft of a thread calls `generate_reply_draft`, which needs `READ_UNIBOX` and `AI_DRAFT`, charges what the dashboard draft charges, and never sends.

Turning it off [#turning-it-off]

Toggle **Draft replies for me** off under **Settings, Workspace**. New replies stop getting drafts immediately. Any drafts already waiting stay until you approve or discard them.

Booking drafts [#booking-drafts]

When [calendar booking](/guides/booking/) is in draft mode, an offer of times, an answer with fresh times, a follow-up or a confirmation arrives as an agent draft like any other, with the block of times kept exactly as offered. A booking draft replaces whatever draft was pending on the thread, so one pending draft per thread always holds. Edit it freely; if the times are removed before sending, the reply still goes out and the booking stops as "a teammate took this over". An answer to someone with a booked call knows about the call.

See also [#see-also]

* [Unified inbox](/guides/unibox/) for the inbox and the manual AI reply draft.
* [AI credits](/guides/ai-credits/) for costs and the balance.
* [AI skills](/guides/ai-skills/) for playbooks the agent follows.
* [Review and approve AI reply drafts](/api/recipes/approve-ai-drafts/) to work the drafts from the API.
* [Replying to cold email replies with AI](/learn/ai-replies-to-cold-email/) for how drafts, auto-reply and booking fit together.
