# Assistant

Chat with the Assistant on its own page, in a panel docked beside any dashboard page, or on a ticket: it investigates escalated tickets, runs your organization's tools with your approval, and builds setup for you.

The Assistant is a chat agent in the dashboard for Organization Admins and Managers, and for department Supervisors. It can read the context of the page you're viewing, gather information about a ticket's requestor from your connected systems, propose resolution plans on escalated tickets, run organization tools, and walk through integration setup conversationally.

To watch it build a runbook, a script tool and an analytics answer, see the [two-and-a-half-minute video](https://www.tryrisotto.com/docs/guides/assistant-video).

The Assistant is available in every organization. Each department has one setting for it, under **Settings → \<department> → Assistant**: **Start automatically when a ticket escalates**. Until September 2026 the feature was called the **Tier 2 Assistant** in Settings; the two names refer to the same feature.

> **Note:** The Assistant is in **Beta** and changes frequently. It is visible only to
> Organization Admins, Managers, and department Supervisors; requestors never
> see its context, plans, or messages.

## Where it appears

You can talk to the Assistant in three places. All three use the same chat, with the same approval cards and approval mode.

### The Assistant page

Click **Assistant** in the sidebar to start a new conversation on a full page. When you send the first message, the conversation gets its own number and URL, such as `dashboard.tryrisotto.com/assistant/ASSIST-42`.

An empty conversation shows starting points: cards such as **Find automation opportunities** and **Summarize ticket trends**. Click a card to send its prompt. Press **/** to start typing. **Enter** sends a message, and **Shift+Enter** adds a new line.

*Product example: A new chat on the Risotto Assistant page, with Assistant highlighted in the sidebar.*

Org Admins and Managers also see [**Artifacts**](https://www.tryrisotto.com/docs/dashboard/artifacts) in the sidebar, the pages the Assistant built. On the Assistant page, the **Assistant settings** menu (the **⋯** button in the header) links to **Assistant settings**, **Sessions** (Org Admins only), and **Usage**.

### The docked panel

On other dashboard pages, open the Assistant in a panel beside the page, so you can keep working on the page while you chat.

1. **Open the panel**

   Click the **Assistant** button at the right end of the page's top bar, or
   press **⌘J** (**Ctrl+J** on Windows). The panel opens on the right, and
   the page moves over to make room. On a ticket page, the button reads
   **Team chat** (see [On a ticket](#on-a-ticket)).

2. **Ask your question**

   Type in the composer, or click a starting point. The starting points match
   the page you're on: on Tickets they offer to triage stale tickets, and on
   Knowledge they offer to review knowledge gaps.

3. **Continue on the full page**

   To move the conversation to the Assistant page, click **Open full screen**
   in the panel header. The conversation continues there.

   *Product example: The Assistant docked beside Tickets, All, with the open launcher in the top bar and Open full screen highlighted.*

The panel header also has **New chat** and **History**, which lists your recent conversations. The top bar button closes the panel too. Drag the panel's left edge to resize it. Press **Esc** or click **Close** to close it. The panel is not available on the Assistant page or in the script studio, which have their own chat.

### On a ticket

On every ticket that has a department, the Assistant button in the top bar reads **Team chat**. The button is there whether or not the department has auto start on. It opens the docked panel on the ticket's team conversation. Each ticket has one team conversation, shared by everyone who can open it. Your team can see it, and the requester can't.

1. **Open the ticket**

   Open the ticket from **Tickets** or your **Inbox**.

2. **Open Team chat**

   Click **Team chat** at the right end of the top bar, or press **⌘J**
   (**Ctrl+J** on Windows). In the Inbox, click the **Assistant** tab next
   to **Ticket History** above the selected ticket instead. It shows the same
   conversation, with the banner "Private to your team — the requestor
   doesn't see this conversation."

   *Product example: A ticket's Team chat open in the docked Assistant panel, with Team chat highlighted in the top bar, the team-chat note and the ticket's starting points.*

3. **Ask about the ticket**

   Click a starting point, such as **Summarize this ticket** or **Suggest next
   steps**, or type your own question. The Assistant knows which ticket you
   mean.

In the panel, switch to the **My chat** tab for your own private conversation, and back to **Team chat** for the ticket's. **New chat**, **History**, and **Open full screen** show only in **My chat**.

### Page context

The Assistant receives the context of the page it's opened from. When you send a message from a ticket page, a chip above the composer shows **Viewing IT-142**, and the Assistant operates on that ticket without you naming it. A chip also shows on a runbook or a script tool. Elsewhere, you can name a ticket by its Risotto key or by the key of the synced ticket in your ticketing system, such as a Jira key.

## Your conversations

On the Assistant page, the header holds the controls for your own conversations.

1. **Open your history**

   Click the **History** icon (**Conversation history**). Your conversations
   are grouped by date: **Today**, **Yesterday**, **Previous 7 days**,
   **Previous 30 days**, and **Older**.

   *Product example: The Assistant's conversation history, grouped by date, with the History button highlighted.*

2. **Search by title**

   Type in **Search conversations** to find a conversation by its title.

3. **Open the conversation**

   Click the conversation to open it.

- **New chat** starts a new conversation. The button shows when a conversation is open. You can also click **Assistant** in the sidebar, or press **⇧⌘O** (**Ctrl+Shift+O** on Windows).
- The pencil icon (**Rename**) beside the title renames the open conversation.
- You can attach an image, such as a screenshot, with the paperclip icon in the composer. Org Admins and Managers can also attach an OpenAPI or Postman spec (`.json`, `.yaml`, or `.yml`).

The panel on the right shows the conversation's **ID** and, for a conversation that started from a ticket or a tool, a link back to it. Under **Actions**:

- **Continue in new chat** starts a new conversation that links back to this one.
- **Archive**, for a conversation that you started, removes the conversation from your history after you confirm. You cannot open it again. Its contents stay in your organization's records.

On a narrow window, click the **Conversation details** icon to open the same panel.

To see the conversations of everyone in your organization, Org Admins use [**All sessions**](#sessions).

## What it does

- **Escalation investigation** — when a ticket escalates in a department with **Start automatically when a ticket escalates** on, the Assistant investigates on its own: it gathers requestor context (People, Identity, or Devices, whichever is connected), states a resolution plan, and executes it with your organization's existing tools, pausing for your approval before each action. Each department can add its own instructions for the investigation phase and for the execution phase (see the [setup guide](https://www.tryrisotto.com/docs/guides/setup-tier-2-assistant#2-add-your-departments-instructions-for-each-phase)). With auto start off, open the ticket's [Team chat](#on-a-ticket) and send a message to start the same investigation by hand. You cannot start the Assistant from a Slack thread. An automatic run always starts in **Read only** mode, whatever the ceiling under **Settings > Assistant > Approval mode**, so it stops at the first action that changes something (see [How tool execution works](#how-tool-execution-works)).
- **Tool execution** — runs your organization's tools (the same tools available to runbooks) as part of any conversation, subject to the approval flow below.
- **Building and setup** — walks through configuration conversationally: creating custom API source tools, authoring [script tools](https://www.tryrisotto.com/docs/integrations/tools/script-tools) end to end (writing, testing, and publishing the code), creating runbooks from a plain-English description, from a standard operating procedure that you paste or that is in your indexed knowledge, or from your template library in the house style, connecting external MCP servers (Linear, Notion, and others) and importing their tools, and building [Artifacts](https://www.tryrisotto.com/docs/dashboard/artifacts) — self-contained dashboard pages wired to your organization's tools.
- **Applications and access rules**: paste a list, for example one row per app with the Okta group, the approvers, and whether a reason is required. The Assistant creates the [applications](https://www.tryrisotto.com/docs/dashboard/applications) and [access rules](https://www.tryrisotto.com/docs/dashboard/applications/access-rules) in batches of up to 100 and asks for your approval before each batch. New access rules are off, so they grant no access until you turn on **Enabled?** on each rule, or ask the Assistant to turn them on. An approver must be an email address or a Slack channel. The Assistant cannot set profile conditions or IdP group approvers. It cannot read an attached spreadsheet or CSV file, so paste the rows into the chat. Only Org Admins and Managers can do this.
- **Questions and analysis** — answers from your Risotto data (including [campaigns](https://www.tryrisotto.com/docs/dashboard/campaigns#tracking-responses) and their responses) and searches your indexed knowledge directly, with tables and structured output where appropriate.
- **Alerts**: ask why something you set up stopped working, for example an index or the ticket sync. The Assistant reads the open error and warning alerts from the last 14 days on the [Alerts page](https://www.tryrisotto.com/docs/dashboard/alerts) and explains a matching alert. Only Org Admins and Managers get this. No matching alert does not prove that the setup works, because some failures raise no alert.
- **Knowledge search**: the Assistant searches your org-wide knowledge and the knowledge of every department where you can read tickets, in one search. Name a department to search it.
- **Tickets for other people**: ask the Assistant to open a ticket for someone else, with their email. It creates the ticket with that person as the requestor, in a department that you can see. In **Read only** mode, it waits for your approval first. You can also ask it to send the requestor a Slack DM about the new ticket. The **New ticket** button on **All tickets** and the **Inbox** opens a new chat that starts this request for you. See [New ticket](https://www.tryrisotto.com/docs/dashboard/tickets/all#new-ticket).
- **Data work and files** — when you ask for data work (for example a report, an export, a CSV, a spreadsheet, a count, a total, or an average), the Assistant can run a script in a sandbox to compute the result instead of reading every record into the chat. Only a script can save a file: it saves the data it computes (`.csv`, `.json`, `.md`, or `.txt`, up to 1 million characters each). All files in one conversation can hold 128 million characters in total. Each save counts toward that total, including a new version of a file that already exists. Text that a tool returns for you to read, such as the knowledge-gaps report, appears in the reply, not as a file. The file appears in the conversation as a card with its size, line count, and a 12-line preview, plus a copy button and **Download**. A `.csv` file previews as a table: the first row is the header, and the table shows the next 12 rows. When the file has more rows, a button at the bottom of the table says how many rows are hidden (for example **40 more rows · Download all 52**) and downloads the file. Files belong to the session they were saved in. A new conversation starts with none. The sandbox is available in every conversation. You do not need a keyword such as "code mode" to turn it on.

### Automations in the Assistant

An Org Admin can ask the Assistant to work with [automations](https://www.tryrisotto.com/docs/dashboard/automations) in their own dashboard conversation. The Assistant can:

- List automations and their runs, and show the details of one automation or one run.
- Create an automation. A new automation starts turned off, so nothing runs until it is turned on.
- Change an automation's name, instructions, schedule, departments, ticket priorities, or approval level.
- Turn an automation on or off.
- Run an automation now. This is a real run against your live systems, also when the automation is off.

Each create, change, turn on or off, and run shows an approval card. The card shows in every approval mode, also in **Auto-accept everything**. If someone changes or deletes the automation while the card is open, Risotto denies the approved call.

The Assistant cannot turn on or run an automation while a tool that the automation names is not published and enabled. It cannot run a webhook automation now.

The Assistant can set the approval level to **Always** or **Destructive changes** only. These actions stay in the dashboard:

- Delete or duplicate an automation.
- Create an automation with a webhook or Slack Connect invite trigger.
- Change the trigger of an existing automation.
- Set the approval level to **Never**.
- Approve the pending tool calls of an automation run.

Other roles get no automation tools. Conversations on a ticket and conversations shared with a department also get none. In those cases, the Assistant can build the tools that the automation calls. Then it links the **New automation** page, where an Org Admin finishes the automation.

## How tool execution works

When a conversation reaches a point where the Assistant needs to run a tool:

1. The Assistant proposes the tool call. A **Tool approval** card shows in the chat with the tool, a one-sentence rationale, and the parameters it will run with. Approvals also appear as action cards on the ticket timeline.
2. Nothing executes until you respond to the card. **Approve** runs the tool with exactly the parameters shown; **Deny** cancels it and the Assistant continues without it.
3. The result returns into the conversation, where a collapsible **tool activity** trace shows each call and its result — status, duration, and expandable details down to the raw parameters. A decided card folds into one line, such as "Approved 2 of 3 actions"; click it to see the details again.

*Product example: A Tool approval card in an Assistant conversation, with Approve highlighted.*

When the Assistant asks for several actions at once, they share one card. Click **Approve** or **Deny** on each row, or use **Approve all** or **Deny all**, then click **Submit decisions**. The button counts your decisions, for example "2 of 3 decided", and only approved actions run.

A tool from a custom API source or an external MCP server shows its source name before the tool name. The tool name links to the tool's page in the dashboard.

### Approval mode

Every conversation starts in **Read only** mode, so every change waits for a card. The approval mode under the composer shows the mode of the open conversation, and you can raise it for that conversation only. A new conversation starts in **Read only** again.

1. **Open the approval mode menu**

   Click the mode under the composer. It reads **Read only** in a new
   conversation.

2. **Choose a mode**

   Click the mode you want:

   - **Read only**: all changes require approval.
   - **Auto-accept changes**: non-destructive changes run without approval.
   - **Auto-accept everything**: destructive tools also run without approval.

   *Product example: The Assistant's Approval mode menu, with Auto-accept changes highlighted and Auto-accept everything not allowed.*

An Org Admin or Manager sets the highest mode operators may choose (see [Set the highest approval mode](#set-the-highest-approval-mode)). A higher mode stays in the menu but cannot be selected, and says "Your organization does not allow this." Above **Read only**, the menu warns that the Assistant may act without showing you first, and that every action is recorded against your account.

A tool set to **Required** approval always shows a card in the Assistant, whatever the mode. On that card, an Org Admin or Manager can click **Approve and stop requiring approval**: it runs the call and removes the tool's **Required** approval for everyone in your organization, for good.

An approval card can also raise the mode. Click **Approve and Enable Auto Mode** to approve the action and switch the conversation to **Auto-accept changes**, or to **Auto-accept everything** for a destructive tool. The button shows only on a card for a single tool call, and only when the org ceiling allows that mode. The rest of the conversation then runs in the new mode.

### Scripts that call tools

A data-work script can also call tools that change things. If the script calls a tool that your approval mode stops, one approval card appears before the script runs. The card lists each of these tools under **This program calls tools that need approval**, with one of these badges:

- **write**: the tool changes something.
- **hard to undo**: the tool makes a change that is hard to reverse.
- **needs approval**: the tool is set to **Required** approval.

The card then shows the full script. **Approve** runs the script once, exactly as written. Nothing runs before you approve, so the card cannot show a preview of the result. A script cannot call a built-in tool set to **Required** approval. The Assistant asks for that tool on its own card.

A script can call a tool from a custom API source or an external MCP server by its alias, `tools.<source>.<tool>` (for example, `tools.linear.get_issue`). A read-only tool from these sources runs without a card. Any other tool from these sources puts the script on the approval card in every approval mode, also when the tool is set to **Required** approval. Risotto treats every tool from an external MCP server as a tool that makes changes, so a script that calls one always shows the card. The card and the tool activity trace show each tool by its own name.

A script can also call an outside API over HTTP, for example to test a vendor API before the Assistant writes a script tool for it. The script declares each domain it calls. It can use a credential from your [Credential Vault](https://www.tryrisotto.com/docs/dashboard/settings/credential-vault), which Risotto adds to the request only when the request leaves the sandbox. If the Assistant needs a credential that is not in the vault, it asks you to add it on a secure credential card. A script that declares a domain or uses a credential always shows the approval card first, in every approval mode, even if it only reads. The card adds these rows:

- **Reaches over HTTP**: each domain, and whether the script makes only GET requests or also POST, PUT, PATCH, and DELETE requests.
- **Uses vault credentials**: each credential the script uses and, for a department credential, the hosts it may reach. Risotto adds the value only to requests to those hosts and their subdomains. See [Department credentials](https://www.tryrisotto.com/docs/dashboard/settings/credential-vault#department-credentials).

## Assistant settings

Org Admins and Managers configure the Assistant for the whole organization under **Settings → Assistant**. The page has three tabs: **Settings**, **All sessions** (Org Admins only), and **Usage**.

### Set the highest approval mode

1. **Open the Assistant settings**

   Go to **Settings → Assistant**. The **Settings** tab opens.

2. **Choose the ceiling**

   In the **Approval mode** card, pick **Highest mode operators may choose**:
   **Read only**, **Auto-accept changes**, or **Auto-accept everything**.

   *Product example: Settings, Assistant: the Web Search and Approval mode cards, with the highest mode highlighted.*

3. **Save**

   Click **Save** at the top of the **Web Search** card. Risotto shows
   "Assistant settings saved".

The same tab has the **Web Search** card. **Enable web search** lets the Assistant search the web. When it is off, the Assistant uses only your organization's knowledge. Under **Allowed domains**, enter one domain per line to limit web search to those domains, or leave it empty to allow all domains.

### Sessions

Every Assistant conversation is a **session**, numbered per organization (`ASSIST-1`, `ASSIST-2`, …). Only Org Admins see the **All sessions** tab. It lists the sessions in your organization, with their first message, owner, number of messages, and creation date, so an admin can review what the Assistant was asked and what it did. An archived session shows a **Deleted** badge.

- Opening **someone else's session** shows a read-only transcript, including every tool call, its rationale, and who approved it.
- Opening **your own session** takes you back to the live chat on the Assistant page, where you can continue it.

### Usage

The **Usage** tab counts the organization's conversations, messages, actions taken, and proactive runs, and how many approvals were approved, denied, expired, or are awaiting a decision. A chart shows conversations per week.

## How credential collection works

Some setup tasks require a secret — an API key or token for a service being connected. The Assistant doesn't accept secrets as chat messages. Instead:

1. It issues a **credential card** in the conversation. The card names the tool, API source, or MCP server the secret is for, so two cards in one conversation are not mixed up.
2. The value you enter goes directly to encrypted credential storage.
3. The transcript records only that a credential was provided — the value itself never appears in the conversation or the Assistant's context.

When the Assistant asks for a vault credential, it can suggest the hosts that the secret may go to. The card fills **Allowed Hosts** with them and warns "Risotto has not checked this host". Type each suggested host in its **Type the host to confirm it** field, or remove it. **Save to vault** stays off until you confirm or remove every suggested host. See [Credential Vault](https://www.tryrisotto.com/docs/dashboard/settings/credential-vault).

## What context it can access

- **Requestor context comes from your connected integrations.** The Assistant looks up the requestor in three data sources: People (your HRIS), Identity (your IdP), and Devices (your MDM). The department's **Assistant** settings page shows which of the three are connected, under **What the Assistant can use**. A department with none connected still gets resolution plans, generated without requestor context.
- **Role-scoped visibility.** The Assistant is available to Org Admins, Managers, and department Supervisors, and its output on tickets is visible only to those roles. An Agent role does not give access. See [Permissions](https://www.tryrisotto.com/docs/operating/permissions).

## The Assistant for department Supervisors

A Supervisor of a department can use the Assistant without an Org Admin or Manager role. The Assistant acts as the Supervisor, with the Supervisor's own access:

- It reads tickets only in the Supervisor's departments. **Team chat** shows on those tickets.
- It creates and edits runbooks and [ticket categories](https://www.tryrisotto.com/docs/dashboard/settings/categories) only for the top-level departments that the Supervisor supervises.
- It runs and tests only the tools of those departments, with the same approval cards.
- It cannot change forms, access rules, applications, tool setup, script tools, custom API sources, external MCP servers, automations, or the department's **Assistant** settings. It tells the Supervisor that an Org Admin manages these.
- It cannot read the organization's alerts.
- In the sidebar, the Supervisor sees **Assistant**, but not **Artifacts**. **Settings → Assistant** and the **⋯** menu on the Assistant page are for Org Admins and Managers.
- The Supervisor can attach images, but not API specs.

Risotto checks the Supervisor role on every message. When the role is removed, the conversation stops.

## Getting started

Auto start is one toggle per department. New departments have it on. There is no organization-wide switch. Only Org Admins and Managers can change it.

1. **Open the department's Assistant settings**

   Go to **Settings → \<department> → Assistant**.

2. **Turn on auto start**

   In the **Auto-start** card, turn on **Start automatically when a ticket
   escalates**. The change saves when you click the switch.

3. **Add instructions**

   Optionally, under **Investigation** and **Execution**, add **Additional
   instructions (optional)** for each phase, up to 2,000 characters each. Click
   **Save**.

More detail is in the [Assistant setup guide](https://www.tryrisotto.com/docs/guides/setup-tier-2-assistant).

Source: https://www.tryrisotto.com/docs/dashboard/assistant
