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

# Review and publish

> Work the review queue, decide on each item, and take a Brain version live.

Review is where your team decides what Kai may say. Publishing is where that
decision reaches customers. This page walks through both.

You need the permission to review, and the permission to publish. An Agent can
read the Brain but cannot change it.

## Work the review queue

<Steps>
  <Step title="Open the queue">
    Go to **Brain**. The band above the six categories states how many items
    wait for a decision.

    <Frame caption="The review band above the categories.">
      <img src="https://mintcdn.com/kai-support/RaOwCZDRMIUnRjWu/images/en/knowledge-review-banner.png?fit=max&auto=format&n=RaOwCZDRMIUnRjWu&q=85&s=7b6a1e4d6a0bedb3227bcd21a614e4af" alt="The review band with the Open the queue button" width="2784" height="260" data-path="images/en/knowledge-review-banner.png" />
    </Frame>

    Select **Open the queue**. The queue holds every undecided item from all
    five reviewable categories, in one list. Publish blockers come first. After
    them, the most frequent situations come first.
  </Step>

  <Step title="Understand the filters">
    Above the list there is a row of filters:

    * **Everything.** All undecided items, in the order described above. This is
      the default and the right one most of the time.
    * **One category.** Answer Library, Conversation Flows, Human Handoff,
      Communication Rules, Brand Voice. Each filter shows its own count. Use one
      when a person on your team owns one category.
    * **Blocks publishing.** Only the items that stop you from publishing this
      version. Use it when you want to go live today.

    When you filter to one category and more than one item is left, an
    **Approve** button appears with a count. It approves the visible items of
    that category only. There is no button that approves the whole Brain in one
    click.

    <Frame caption="Two queue rows. The badge marks an item that blocks publishing.">
      <img src="https://mintcdn.com/kai-support/RaOwCZDRMIUnRjWu/images/en/review-queue-item.png?fit=max&auto=format&n=RaOwCZDRMIUnRjWu&q=85&s=16d2486eab1c38f20af5c96807698dc5" alt="Two review queue rows, one badged Blocks publishing" width="1700" height="360" data-path="images/en/review-queue-item.png" />
    </Frame>
  </Step>

  <Step title="Read one item">
    Select a row. The right side shows the whole item and the real conversations
    it came from.

    <Frame caption="The review queue: the list on the left, the item on the right.">
      <img src="https://mintcdn.com/kai-support/RaOwCZDRMIUnRjWu/images/en/review-queue.png?fit=max&auto=format&n=RaOwCZDRMIUnRjWu&q=85&s=d1b91f923a30c341596148b88931b53f" alt="The review queue with the list and the editor" width="2880" height="2000" data-path="images/en/review-queue.png" />
    </Frame>

    Read the examples under **Examples from past conversations** before you
    decide. They are the messages your own team sent. If the proposed answer
    does not match them, the answer is wrong, not the examples.
  </Step>

  <Step title="Read the count next to the item">
    Each row ends with a number, for example `×118`. It is the number of past
    conversations that produced this item. For an answer, it is how many
    separate customers asked that question.

    Treat a high count as a warning to slow down, not as a reason to trust the
    item. A high count means two things at once: the item is well attested, and
    it will be used often. An error in an item with `×118` behind it reaches
    many customers. An error in an item with `×2` behind it reaches almost
    nobody.
  </Step>

  <Step title="Decide">
    You have three decisions.

    * **Approve.** The text is correct as written. Select **Approve**.
    * **Edit.** Change the text in the form, then select **Save and approve**.
      The item becomes **Edited**, and your words are what Kai uses.
    * **Reject.** Kai must not use this at all. Select **Reject**. Kai stops
      retrieving the item. It never reaches a reply again.

    The decided item leaves the queue and Kai opens the next one. To move
    without deciding, use `j` and `k`. To decide from the keyboard, use `a` to
    approve and `r` to reject.
  </Step>
</Steps>

<Tip>
  Edit rather than reject when the topic is right and the wording is wrong.
  Reject only when Kai must never answer that topic.
</Tip>

## Publish

Publishing points every connected channel at this version, and freezes it.

### What blocks publishing

Kai blocks a publish only for content that could reach a customer without a
human reading it first:

* the **Brand Voice**, which must exist and must be reviewed;
* **every Human Handoff rule**, without exception;
* **every Conversation Flow that is eligible for autopilot**;
* **every Answer Library entry that answers an intent one of those flows
  covers**.

Everything else can stay unreviewed. That is on purpose. An unreviewed answer
in a published Brain still writes drafts in copilot, where a human reads every
word before the customer does. If unreviewed content blocked a publish, no team
would ever get to review its long tail against real traffic.

<Note>
  The rule that protects you is applied per reply, not per Brain. On autopilot
  with the scope **Anything a reviewer approved**, Kai sends a reply on its own
  only when the answers it used and the flow it followed are approved or edited.
  A reply built from unreviewed content becomes a draft.
</Note>

### Publish the version

<Steps>
  <Step title="Clear the blockers">
    In the queue, select the **Blocks publishing** filter. Decide every item in
    it. The count in the header goes to zero.
  </Step>

  <Step title="Select Publish">
    Go back to **Brain** and select **Publish** in the header.
  </Step>

  <Step title="Read the confirmation">
    Kai states the consequence before it happens: every channel answers from
    this version immediately, and the version stops being editable.

    If unreviewed entries are still in the version, Kai names how many. They
    ship with the version, and they can only draft for a human.
  </Step>

  <Step title="Confirm">
    The previous published version becomes **Archived**. It stays readable.
  </Step>
</Steps>

### What publishing freezes

After a publish, the version is read-only. You cannot approve, edit or reject
anything inside it. Facts are the exception: current information is not part of
a version, so a price change still takes effect on the next reply. Read
[Current information](/product/current-information).

### Edit as a new draft

To change a published version, open it and select **Edit as a new draft**. Kai:

* creates a copy, named `v1.1`, `v1.2`, and so on;
* carries every approval, edit and rejection across to the copy;
* leaves the published version untouched and serving customers.

Work in the copy, then publish it. If you select **Edit as a new draft** twice,
Kai opens the draft that already exists. It does not create a second one.

## Split the work across a team

The queue is one list, but the filters make it several.

* **Give each category an owner.** Human Handoff and Brand Voice decide when a
  machine talks to a customer unsupervised, so give them to the person who
  answers for that. Answer Library is the long tail, and it is the part that can
  be shared widely.
* **Clear the blockers first, together.** Use the **Blocks publishing** filter
  in one sitting. It is a short list, and it is the only list that stops you
  from going live.
* **Two people can work at the same time.** Each decision is written on its own.
  A reviewer never overwrites another reviewer's decision.
* **Leave the tail for later.** Go live in copilot with the tail unreviewed.
  Your team then reviews real drafts in the inbox instead of guessing which
  answers matter.

## Next

<CardGroup cols={2}>
  <Card title="The knowledge base" icon="book-open" href="/product/knowledge-base">
    What each category holds, and what each review state means.
  </Card>

  <Card title="Copilot, autopilot and off" icon="sliders" href="/product/modes">
    A safe order to turn each mode on after a publish.
  </Card>

  <Card title="Current information" icon="tag" href="/product/current-information">
    The values you can change without a publish.
  </Card>

  <Card title="Retrain the Brain" icon="rotate" href="/product/retrain">
    Compile a new draft, and keep the decisions you already made.
  </Card>
</CardGroup>
