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

# AI Assistant

> Ask questions about your community in plain language, right inside Studio

> [!NOTE]
>
> The AI Assistant is in **beta**. It is off by default. Contact your Tightknit
> representative to enable it for your community.

The AI Assistant answers questions about your community without you having to build a
filter or export a report. Ask it in plain language, and it looks up the answer using the
same data you can already see in Studio.

It knows which community you're in and which page you're looking at, so you can ask about
"these members" or "this event" without spelling out what you mean.

## Opening the assistant

There are two ways in, and both open the same panel on the right side of Studio:

* Click the **sparkles icon** in the Studio header, next to your profile menu.
* Press <kbd>Cmd</kbd>+<kbd>K</kbd> (<kbd>Ctrl</kbd>+<kbd>K</kbd> on Windows) and select
  **Ask AI Assistant**.

The panel stays open as you move around Studio, and your conversation is kept while you
navigate or reload the page. Press <kbd>Esc</kbd> to close it, and use the
**New conversation** button (the pencil icon) in the panel header to start a fresh
conversation.

## What it can do

The assistant can read anything you already have permission to see, including:

* **Members and groups** — find members, check group membership, look up profiles
* **Posts, comments, and discussions** — search content, read threads, check what was published
* **Events** — list events, check attendees and RSVPs
* **Journeys, awards, and milestones** — check progress and assignments
* **Feeds, channels, and tags** — see how content is organized
* **Site, branding, and content collections** — check configuration
* **Activity history** — see what has happened in your community
* **Tightknit documentation** — answer questions about how a feature works, with a link
  to the relevant page

When an answer draws on the documentation, a **Sources** list appears beneath it with the
pages the assistant read. That list is built from what it actually looked up, not from
what it wrote. If you ask how a feature works and no Sources appear, it answered from
memory rather than checking, and is worth verifying.

Good questions to start with:

* "Which members joined this week?"
* "What were the most active discussions last month?"
* "Who hasn't completed the onboarding journey?"
* "How do I set up custom fonts?"

## Getting you where you're going

As well as answering, the assistant can move you around Studio:

* **Open a page.** Ask it to "open the members page" or "show me that event" and it takes
  you there, rather than giving you a link to click.
* **Open a create dialog.** Ask it to "create an event" and it opens the same Create Event
  dialog you'd get from <kbd>Cmd</kbd>+<kbd>K</kbd>, ready for you to fill in.

It only offers pages and dialogs your own role can already open: the list is exactly what
your sidebar and command menu show you. It does one of these per reply, so it can't send
you somewhere unexpected mid-answer.

## Making changes

The assistant can make changes for you. It always shows you the change first, and makes it
only after you approve it.

When the assistant wants to make a change, a card appears in the conversation showing the
details, with **Approve** and **Decline** buttons:

* **Approve** makes the change, and the card shows the result.
* **Decline** cancels it. Nothing changes, and the assistant won't try to make the change
  another way unless you ask it to.

The card names each record the change touches, as the assistant looked it up in this
conversation, so check those names before you approve. A record marked **Not looked up in
this conversation** shows only its ID: ask the assistant to look it up, or check the record
in Studio first.

Deleting something can't be undone, so a delete's card is red and says so.

A change you don't approve within 30 minutes expires, and its card is marked **Expired**.
The records it named may have changed since it was proposed, so the assistant won't make
it. Ask again if you still want the change.

It only offers changes your role already has permission to make. Tightknit keeps an audit
record of every change the assistant makes, including who approved it.

If you'd rather fill in the details yourself, ask it to open the create dialog instead.

## Saved conversations

The assistant saves your conversations, so you can close the panel or reload Studio
without losing them. A conversation is saved once you send its first message.

When you start a new conversation, your five most recent saved conversations are listed
under the suggested prompts. Click one to pick up where you left off, or click **View all
history** to see the rest.

Click the **Conversation history** button (the clock icon) at the top of the panel to see
all your saved conversations in this community. From there:

* **Reopen a conversation** — click it to pick up where you left off.
* **Rename a conversation** — open its **⋯** menu and choose **Rename**. Until you rename
  it, a conversation is named after your first message.
* **Delete a conversation** — open its **⋯** menu and choose **Delete**. A deleted
  conversation can't be recovered.

What is kept, and for how long:

* **Only you can see your conversations.** Other admins in your community can't list or
  open them.
* **Each community keeps its own.** Conversations you have in one community don't appear
  in another.
* **Your 25 most recent are kept.** Starting another conversation in a community that
  already has 25 deletes your oldest one there.
* **A conversation is deleted 30 days after its last message.** Sending a new message
  restarts the 30 days. Opening or renaming it doesn't.

### When a conversation gets long

A conversation can only hold so much. As it gets close, a notice above the message box
says it's **almost full**, and the assistant's replies may be shorter from then on. Once
it's full, you can still read it in your history, but you'll need to start a new
conversation to keep going.

## What it can't do

> [!WARNING]
>
> The assistant can't make any change you haven't approved, deletes included. The dialogs
> it opens for you stay yours to fill in and save.

Other current limitations:

* **One community at a time.** The assistant only sees the community you're currently in,
  and a conversation stays in the community it started in.
* **Long conversations fill up.** A very long conversation, or one with several large
  lookups, eventually stops accepting new messages. Start a new conversation to keep
  going. See [When a conversation gets long](#when-a-conversation-gets-long).
* **No images or files.** The assistant can't upload an image or file, so change a logo,
  banner, or cover image on its Studio page.
* **It can be wrong.** Like any AI assistant, it can misread a question or summarize
  imprecisely. For anything you're going to act on, check the underlying Studio page.
* **Usage is rate limited** to protect your community's performance.

## Who can use it

Access requires two things:

1. The feature must be **enabled for your community**.
2. Your Studio role must include the **`studio:assistant:use`** permission. Admins have it
   automatically; other roles need it granted. See [Access & Roles](https://docs.tightknit.ai/studio/access.md).

If either is missing, the sparkles icon and the Cmd+K entry don't appear. The assistant
also becomes unavailable if a workspace's billing is suspended.

Everything the assistant does is checked against **your** permissions, not a shared
service account. If your role can't see a section of Studio, the assistant can't see it
for you either. A member manager gets member data, but not billing.

## How your data is handled

> [!WARNING]
>
> Answering your questions involves sending community data to **OpenAI**, our AI provider,
> through Tightknit's account. The details below describe exactly what is and isn't sent.

**Email addresses are redacted.** Before any community data reaches the AI provider, email
addresses are replaced with `[REDACTED_EMAIL]`. This means the assistant cannot tell you a
member's email address. You can still find it on the member's page in Studio.

**Record IDs are sent as-is.** Identifiers like `prfl_Ab3dEf6hIj9k` are not redacted,
because the assistant needs them to follow one lookup with another — finding a member and
then checking their group membership, for example. These IDs are meaningless outside your
community and are not personal information on their own.

**Other community content is sent as needed.** When you ask about posts, discussions, or
profiles, the relevant content is sent so the assistant can answer. It only ever retrieves
data you already have permission to view.

**Community data is treated as information, not instructions.** Content written by your
members is clearly separated from the assistant's instructions, so a message containing
something like "ignore your instructions" is read as text to report, not a command to
follow.

**Nothing is sent until you ask.** The assistant is idle until you send a message; it does
not read your community in the background.

**Conversations are stored for 30 days after their last message.** Tightknit stores your
conversations, including the community data the assistant looked up while answering, so
you can come back to them. Each one is deleted 30 days after its last message, when you
delete it, or when it falls outside your 25 most recent. See
[Saved conversations](#saved-conversations).

If your organization has specific requirements about what may be sent to an AI provider,
talk to your Tightknit representative before enabling the assistant.

## Giving feedback

The assistant is in beta and we're actively improving it. If it gets something wrong,
misses data you expected, or you'd like it to do something it can't yet, let your
Tightknit representative know. Including the question you asked helps us a lot.


## Related topics

- [AI at Tightknit](https://docs.tightknit.ai/ai/overview.md)
- [Tightknit Data Security & Privacy](https://docs.tightknit.ai/security/data.md)
- [AI Moderation](https://docs.tightknit.ai/moderation/ai-moderation.md)
- [AI & Agent Settings](https://docs.tightknit.ai/channels-and-feeds/agent-settings.md)
- [Getting Started with Tags](https://docs.tightknit.ai/tags/getting-started.md)


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