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

# Assist

Assist is the assistant built into your Siit admin dashboard. Ask it a question about Siit and it answers from the documentation. Ask it to do the work and it builds the configuration for you

{% embed url="<https://youtu.be/SjUdPAV7KU0?si=afQu3bn8o-cnNLby>" %}

#### ► Assist can only see and do what you can

This is the whole design. Assist runs on your behalf, with your role and your permissions. It cannot read a request, a person, or an inbox you cannot already open in the dashboard, and it cannot perform a write you could not perform yourself in the UI. Scoping is identical to the Siit MCP.

The one way past your Siit workspace is an MCP connector you set up yourself, and it reaches only the tools and the account you authenticated with. See MCP connectors below.

On top of that, every action that changes your workspace pauses and asks you first. You approve or reject it in the conversation before anything is written. A resolution plan is approved or denied as a whole.

#### What you can do

Assist opens on three starting points, each with ready-made prompts.

|                      | What it does                                                                                          | Ask it                                                                                                                                                            |
| -------------------- | ----------------------------------------------------------------------------------------------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| **Build my setup**   | Creates and edits your configuration: services, service forms, workflows, inboxes, knowledge articles | "Suggest three workflows I should set up based on my requests." / "Create a service catalog based on past requests." / "Create an onboarding workflow."           |
| **Ask the docs**     | Answers product and setup questions from Siit's public documentation, Help Center, and API reference  | "Where do I set up my app access policies?" / "How do I set up SSO?" / "How do SLA rules handle business hours?"                                                  |
| **Analyze requests** | Reads your live request data and turns it into answers, summaries, and drafts                         | "How many requests last week, by service?" / "Find open requests older than 7 days with no assignee." / "Write my inbox weekly update from last week's requests." |

Assist reaches your workspace through the same toolset as the Siit MCP, so anything the MCP can read or write, Assist can too: requests, people, apps and equipment, services, workflows, app access roles, knowledge base articles, and your organization structure. See MCP for the full tool list.

On top of those three, Assist can take a request off your hands end to end — see [Resolve a request with Assist](#resolve-a-request-with-assist).

### What to ask first

Everything below runs on your real workspace. Each prompt is a starting point, not a fixed command: refine it, redirect it, or push further in the same conversation.

### Build the workflow you never got to

You know some requests deserve a workflow. Building one means finding the time, then clicking through it.

{% embed url="<https://youtu.be/47EgrtC9gXE?si=RkTEgZbB6ShRm4aF>" %}

**What you get:** the gaps in your real activity, ranked, then the whole workflow drafted in the same conversation, triggers, steps, and approvals included.

You can also start from the Workflows page: **New** opens a prompt, you describe the workflow you want, and Assist builds it and opens it — see [Starting a new workflow](/workflow/triggers-overview.md#starting-a-new-workflow).

Workflows Assist creates start as drafts. Review them on the Workflows page and click **Set live** when they are ready — see [Workflow status](/workflow/triggers-overview.md#workflow-status).

### Write the article everyone keeps asking for

Your help center answers the questions you thought of. The gaps only show up in the requests.

{% embed url="<https://youtu.be/larNM7GfD5s?si=QBZp9yp59fWienI9>" %}

**What you get:** the gap in your knowledge base, found in your real request history, and the article written to close it. You review, approve, and it is published.

### Rebuild your old setup from a screenshot or an export

Moving from another tool usually means recreating every form and service by hand.

{% embed url="<https://youtu.be/W-PTXnnsDvw?si=jffUZawm9ahBVesD>" %}

**What you get:** your forms recreated as Siit services, and a catalog drafted from your real request history, ready for your approval.

### Let Friday write itself

The weekly update means re-reading a week of requests to summarize it.

{% embed url="<https://youtu.be/fc4-SjAe3Ow?si=mS83WNxcjKMFcnVe>" %}

**What you get:** your update drafted from live request data, ready to edit and send.

### Get answers where you work

Setup questions usually mean searching the help center, or asking support and waiting.

{% embed url="<https://youtu.be/Y0c2Vi3v13w?si=4NRwUDQcP6VvD8CT>" %}

**What you get:** the answer, drawn from Siit's documentation, without leaving what you were doing.

***

#### How it works

1. Open **Assist** from the **Ask Assist** button in the top bar of the admin dashboard, or double-tap <kbd>Ctrl</kbd> from anywhere in the console — on both Mac and Windows, and even while you are typing. It opens in a side panel with the cursor already in the composer, and the page behind it stays usable. Double-tap <kbd>Ctrl</kbd> again to close it, which leaves your focus where it was.
2. Ask in plain language, or pick one of the suggested prompts.
3. Assist works out loud. You see each step as it reads your workspace, checks the documentation, and prepares changes.
4. Before any write, Assist stops and asks for approval. The conversation stays paused until you approve or reject.
5. Stop a run at any point with the stop button. Nothing already approved is rolled back.
6. Long conversations keep going. When a thread outgrows what the model can hold, Assist compacts it by itself: you see **Compacting…**, then **Compacted context**, and the answer continues from a summary of everything before that point.
7. Conversations are saved. Reopen a past one from the history icon to pick up where you left off.

Assist links the records it names. When an answer mentions one of your requests, people or services, the name is a link straight to that record in the dashboard.

Anything Assist did on your behalf is labelled **via Assist** next to its date, wherever the activity shows up: the request feed, a group's timeline, and the People, Apps, Equipment and app access timelines. The actor stays you — the label only says which route the change took.

#### Resolve a request with Assist

Assist can work a whole request, not just answer questions about it.

Open the request and click the Assist icon in the message composer, or pick **Solve this request with Assist** from the composer's **⋯** menu. The icon is pinned to the composer by default, alongside the article, file, emoji and message template actions; unpin it from the same menu if you would rather not see it. Assist opens in its side panel, already prompted with that request's number.

Assist then reads the request and proposes a **resolution plan**: a numbered list of every action that would change something in your workspace or your connected tools, each with a short sentence naming what it does, in the order it would run. A plan often ends with a message to the requester explaining what was done; you can read that message in full, and copy it, before you decide anything.

You decide the plan as a whole:

* **Apply** runs the actions in order. The plan stays on screen while they run, and each step ends up labelled **Done**, **Failed** or **Not run**. The run stops at the first failure, so later steps stay **Not run**.
* **Deny** runs nothing. The plan is kept in the conversation, marked **Denied**, so you can see what was proposed.

A step the run was interrupted on is labelled **Unconfirmed**: it may or may not have been applied, so check its effect before running it again. Until you decide, the conversation stays paused and the composer is blocked — Assist will not take a new instruction with a decision outstanding.

Reading is never gated this way. Assist looks the request up, checks the person, the device and your configuration on its own; only the changes wait for you.

{% hint style="info" %}
The resolution actions only show on a request that is still open. Once the request is resolved, Assist no longer offers to work it.
{% endhint %}

#### Where its answers come from

Assist uses these sources:

* **Your workspace**, for anything about your own requests, people, and configuration.
* **Siit's public knowledge**, for anything about how the product works: [docs.siit.io](https://docs.siit.io), [help.siit.io](https://help.siit.io), and the API reference at [developer.siit.io](https://developer.siit.io).
* **The API documentation of your custom integrations**, when you gave one an **API documentation URL**. Assist reads those pages the same way it reads Siit's own, so it can work out the endpoint, the method and the payload a [custom action](/integrations/custom-integrations.md) needs instead of asking you to paste them. It only follows pages on that same site, and only for integrations your role can see.

Nothing from your workspace is sent to the documentation, and nothing from the documentation is treated as configuration.

#### Attachments

Drop a file or a screenshot into the conversation and Assist reads it. The common case is migration: paste a screenshot of your current setup in another tool and ask Assist to build the equivalent in Siit.

#### Common workflows

* **Trial day one** : "Create a service catalog based on my past requests" turns an empty workspace into a working catalog in one prompt.
* **Migration** : screenshot your existing intake form from your old tool and have Assist rebuild it as a Siit service with the same fields.
* **Find the automation gap** : "Which services get the most requests but no automation?" then have Assist draft the workflow it just recommended.
* **Weekly reporting** : "Write my inbox weekly update from last week's requests."
* **Setup questions without a support ticket** : "How do SLA rules handle business hours?" answered from the docs, in context, in the product.

#### Tips

* One task per conversation. Assist works from the whole thread, so a conversation that has drifted across three topics is slower and less accurate than a fresh one, even once it has been compacted.
* Say which object you mean. "Update the Onboarding workflow" beats "update the workflow."
* Read the approval prompt before approving. It names the exact action and its parameters.
* When Assist is wrong about the product, it is usually because the documentation is wrong. That is worth reporting.

#### Tell Assist how you work

Assist takes two layers of written guidance. Both are plain language, up to 10,000 characters each, and both apply to every conversation.

* **Workspace guidance**, in **Settings → AI & Agents → Assist**, applies to every admin in the workspace. Use it for the things Assist cannot infer: how your teams are organised, your internal vocabulary, and what it should never touch. For example: *"We run Okta and Jamf. Route hardware requests to IT Ops. Never suggest changes to SLA rules."* Editing it needs the **Can manage Assist** permission.
* **Personal instructions**, in **Settings → Personal → Assist personalization** — also reachable straight from your avatar menu — apply to your own conversations only. Use them for tone, format, and language. For example: *"Keep drafts under three sentences. Write in French. Link the help article when one exists."* Every admin sets their own, with no permission needed.

When both are set, Assist follows the workspace guidance and layers your personal instructions on top of it.

#### Connect your own tools with MCP connectors

Assist can reach beyond Siit through MCP connectors, so it can pull context from the other tools you work in.

An Owner or an admin with **Can manage Assist** turns this on with the **Allow external MCP** toggle in **Settings → AI & Agents → Assist**. While it is off, Assist only runs Siit-managed connectors.

Once it is on, each admin manages their own connectors in **Settings → Personal → Assist personalization → Connect external MCP**. You can also get there from Assist itself: **Connect** on the **Connect your tools for better answers** banner that sits above the composer on the new-conversation screen, and **Manage connectors** in the conversation menu.

{% stepper %}
{% step %}

### Pick or add a connector

The page lists your connectors in two sections: **Your MCP servers**, the servers you added yourself, and **Connectors**, the ones your workspace allows. Filter the page with **All**, **Connected** or **Not connected**, or type in **Search connectors** to find one by name. Nothing matching is not a dead end: any MCP server works here, even one outside the catalog. A server you add yourself needs a name and an `https://` server URL on a public host.
{% endstep %}

{% step %}

### Authenticate

Choose **OAuth**, a **Bearer token**, or **No authentication**, depending on what the server expects. OAuth sends you to the provider's own sign-in page. Servers that need their own headers take up to 20 custom HTTP headers.
{% endstep %}

{% step %}

### Choose the tools Assist can use

Siit discovers the tools the server exposes and lists them under **Tool permissions**. Tick only the ones Assist should be able to call. With none selected, Assist ignores the connector.
{% endstep %}
{% endstepper %}

A connector that is connected shows an on/off switch, **Let Assist use this connector**, on its row and at the top of its **Configure** page. Turn it off to keep the connector set up while Assist stops calling it, and back on when you want it again; Siit confirms either way. A connector that is not connected shows its status instead — **Not connected**, or **Error** when the last attempt failed.

Each row's **⋯** menu carries what you can do with it:

* **Configure** — the server URL, the credentials and the tool selection. This is where you look after an **Error**.
* **Reconnect** — for OAuth connectors, sends you back through the provider's sign-in.
* **Disconnect** — for a connector from your workspace catalog: its credentials and tool selection are removed and Assist stops using it, but it stays in the list ready to connect again.
* **Delete** — for a server you added yourself, which removes it from the list entirely.

Credentials are yours alone: they are stored encrypted against your own account and are never shared with other admins. A personal connector carries its own credentials only and cannot reference your workspace secrets. Disconnecting or deleting a connector deletes its credentials with it.

#### Availability

Assist is available for every admin, on every plan, trials included. There is no AI add-on and no per-seat AI upgrade.

Usage runs on fair use rather than a conversation count. Your workspace runs one shared cycle that resets weekly, and every admin gets their own allowance inside it. Pro workspaces get a larger allowance per admin than trial, Essentials, and Standard.

Assist warns you in the conversation when you are approaching your limit, and again when you have reached it. If you reach it, Assist pauses until the next reset, and the [Siit MCP](https://docs.siit.io/ai/mcp) keeps you going in the meantime.

**Assist usage**

Admins with the **Can view Assist usage** permission can follow consumption in **Settings → AI & Agents → Assist usage**. The table lists every admin with the share of their allowance they have used in the period and when they last used Assist, sortable on either column. Use the arrows to step back through earlier periods, and read **Next reset** for the date the current one rolls over.


---

# Agent Instructions
This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com.

## Querying This Documentation
If you need additional information that is not directly available in this page, you can query the documentation by asking a question.

Perform an HTTP GET request on the following URL with the `ask` and `goal` query parameters:

```
GET https://docs.siit.io/ai/assist.md?ask=<question>&goal=<user_goal>
```

`ask` is the immediate question: it should be specific, self-contained, and written in natural language.
`goal` is what the user is ultimately trying to achieve, the reason they need the answer. Sharing it helps GitBook give you a better, more relevant answer. A goal is most helpful when it describes the outcome the user wants rather than restating the question. For example, with `ask=how do I create an API token`, a goal like `automate deployments from our CI pipeline` lets GitBook tailor the answer to that use case.

The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.

Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.
