> For the complete documentation index, see [llms.txt](https://support.suptask.com/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://support.suptask.com/suptask-ai/ai-assistants.md).

# AI Assistants

Assistants empowered by Agentic AI

Suptask AI works through **Assistants** - configurable helpers that work on your tickets. An Assistant can answer common questions from your own knowledge and take action in the tools your team uses, resolving routine requests so your agents don't have to.

Each Inbox can have one or more Assistants, each set up for the kind of work that Inbox handles.

{% hint style="warning" %}
AI Assistants are a part of the Custom plan and available in beta. [Contact Suptask](/getting-started/contact-support.md) to get started.
{% endhint %}

***

### What an Assistant does

An Assistant helps in two main ways:

* **Answers requests** - it finds the answer in your knowledge and replies, either as a suggestion for an agent to send or directly to the requester. See [Auto-answer requests.](/suptask-ai/auto-answer-requests.md)
* **Takes action** - it looks things up and makes changes in your connected systems, from a single lookup to a task that spans several steps.

An Assistant can drive the conversation with the requester on its own, completing as much of a request as it can before handing over to an agent.

<figure><img src="https://621243658-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FO5JbdwXcuzsmDvD73Ccf%2Fuploads%2Fgit-blob-a2e04b074cc2b1c071ce999cecee82a280217139%2Fimage.png?alt=media" alt="" width="563"><figcaption><p>Agents can use AI assistants as their own co-workers</p></figcaption></figure>

***

### Multi-step workflows

Some requests take more than one step, and often more than one system. Access requests, onboarding and offboarding all follow the same shape: look something up, check what already exists, then make a change.

An Assistant chains those steps together and runs them from the ticket. Unlike a fixed [Automation](/integrations-automations-and-workflows/how-to-set-up-automation-rules.md) rule, it adapts to the specifics of each request rather than following one predefined path, and it reports back on the ticket as it goes.

You write these steps as **workflow tasks** - a numbered list of instructions the Assistant works through in order. Simple Assistants need none: an Assistant that answers questions from your documentation works from its base instructions alone.

***

### Connectors, tools and permissions

You decide exactly what each Assistant can reach and how far it can go on its own.

An Assistant is given one or more **connectors** - the systems it can reach. Each connector provides **tools**, the individual things it can look up or change. You enable only the tools an Assistant should use, and set a mode for each one:

| Mode     | What happens                                                                        |
| -------- | ----------------------------------------------------------------------------------- |
| **Off**  | The Assistant can't use the tool at all.                                            |
| **Ask**  | The Assistant proposes the action and an agent approves it before anything changes. |
| **Auto** | The Assistant runs the tool on its own.                                             |

<figure><img src="https://621243658-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FO5JbdwXcuzsmDvD73Ccf%2Fuploads%2Fgit-blob-72fa4c623b1b230df8860b33850f9a0996fa635e%2Fimage.png?alt=media" alt="" width="375"><figcaption></figcaption></figure>

A safe starting point is every lookup on **Auto**, every tool that changes data on **Ask**, and anything destructive on **Off**. That keeps routine work fast while sensitive changes stay under human control. You can loosen it once you've seen how the Assistant behaves on real tickets.

For requests that need formal sign-off, you can combine this with [Approvals](/working-with-tickets/for-agents-who-respond-and-manage-tickets/how-to-send-and-manage-approvals.md).

{% hint style="info" %}
Permissions belong on the tools, not in the Assistant's instructions. There's no need to write "ask before changing anything" into a prompt - the mode you set does that.
{% endhint %}

***

### When an Assistant runs

An Assistant can start working on a ticket in two ways:

* **Automatically** - on a trigger, such as when a ticket is created. You can limit it to specific Forms, so it only runs on the request types it was built for. Where several Assistants could match, Suptask runs the one whose criteria fit best.
* **Manually** - mention `@Suptask` on a ticket and an agent triggers it themselves.

Results are visible to the agent by default. An Assistant only replies to the requester when you turn that on - see [Auto-answer requests](https://claude.ai/suptask-ai/auto-answer-requests.md).

***

### Where an Assistant posts

An Assistant works inside a ticket, and it can post in more than one place. The Slack connector is built in, so this works without connecting anything.

<table><thead><tr><th width="255.07421875">Tool</th><th>Where it posts</th></tr></thead><tbody><tr><td><strong>Post Message Requester</strong></td><td>The requester's thread - what the person who raised the ticket sees.</td></tr><tr><td><strong>Post Message Responder</strong></td><td>The responder thread - visible to agents only.</td></tr><tr><td><strong>Post Message User</strong></td><td>A direct message to a specific Slack user.</td></tr></tbody></table>

By default an Assistant reports to agents and you do not need to enable the Slack connector for this. Posting to the requester's thread is how it answers the person who asked, and how it brings someone else into the conversation.

An Assistant can also tag a Slack user in what it posts, which sends them a normal Slack notification. That's how a request reaches the right person without being reassigned - useful when the owner of a system isn't an agent in the Inbox at all.

{% hint style="info" %}
Tagging works with a Slack user ID. Have the Assistant look it up with Get User Info, or store the handles in the document it reads.
{% endhint %}

***

### Set up an Assistant

Assistants are set up per Inbox in the Suptask Web app, under **Intelligence**. Adding one takes four steps:

1. **Connectors** - pick the systems the Assistant can reach, enable the tools it needs, and set each tool to Off, Ask or Auto.
2. **Tasks & Personality** - name the Assistant, write its base instructions, choose a tone, and add workflow tasks if the job needs several steps.
3. **Activation** - choose how it's triggered, and which Forms it applies to.
4. **Review & Deploy** - check the summary and deploy.

An Assistant stays **Inactive** until you deploy it, so you can build and review the whole setup before it touches a ticket.

{% hint style="warning" %}
Suptask AI specialists are available to help you set up your first Assistants. [Contact Suptask](/getting-started/contact-support.md) to get started.
{% endhint %}

***

### Related articles

{% content-ref url="/pages/LytgedaqKMhfYKBQMnyQ" %}
[Auto-answer requests](/suptask-ai/auto-answer-requests.md)
{% endcontent-ref %}