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

# Ask Analysts

**Ask Analysts** is Supaboard's conversational interface. Type a question in plain English and your agent queries your data, then returns charts, tables, KPIs, and written explanations — no SQL required.

Ask Analysts is powered by your workspace's [Agents](./agents). Each agent defines what data it can reach and how it interprets your questions.

> Some older screens still call this **Ask Stella**, and the URL is `/ask-stella`. It's the same feature.

# Opening Ask Analysts

Click **Ask Analysts** in the sidebar. If you have an existing conversation, it opens where you left off.

## The empty state

A new conversation offers shortcuts that configure the input for you:

| Card / pill             | What it does                                               |
| ----------------------- | ---------------------------------------------------------- |
| **Upload Files**        | Opens the file picker so you can add context before asking |
| **Go Deeper-Deep Dive** | Switches the mode selector to Deep Dive                    |
| **Try Voice Mode**      | Starts voice input                                         |
| **Connect MCP Tools**   | Opens the tool picker                                      |

Suggested questions are shown below, tailored to your connected data.

# Conversations

## The conversation list

Open the conversations panel from the **topbar**, beside the page title.

Conversations are grouped by the **date they were created** — headings read like *"7th August, 2026"* — newest first. **Search** filters by conversation name. Conversations load 15 at a time; scroll to load more.

Hover any conversation for a three-dot menu with **Rename**, **Control Access**, and **Delete**.

## Starting a new conversation

Click the **New Conversation** button in the topbar.

## Choosing an agent

The agent selector sits in the input row and is pre-filled with your workspace's default agent. You can change it on any turn, not just the first — so you can start with one agent and switch mid-conversation if another has the right data scope.

## Conversation limit

Each conversation supports up to **20 exchanges**. At the limit, the input is replaced by a card titled *"To continue, please start a new conversation"* with a **Continue in New Chat** button. Context from the current conversation carries over into the new one, so you don't lose your place.

# Asking a Question

Type in the input bar at the bottom — *"Ask anything about your data..."* — and press Enter.

## Query modes

Two pills in the input bar control how your question is processed:

### Concise

*Precise to the point answer.* Quick follow-ups and simple lookups where speed matters more than depth. Best for summarised insights, KPIs, and "what" questions.

### Deep Dive

*Deep analytical thinking.* Questions that need real analysis — charts, tables, KPIs, with exploratory insight. Best for "why" and "how" questions.

## Voice input

Click the microphone beside Send to dictate your question. While listening, the input shows a live transcript and a waveform. Click again to stop.

## Attachments and tools

Both live behind the **+** button (*"Add File(s) or Tool(s)"*).

### Add File(s)

Attach a file to give the agent extra context.

| Detail             | Value                                                                           |
| ------------------ | ------------------------------------------------------------------------------- |
| **Size limit**     | 5 MB per file                                                                   |
| **Accepted types** | PDF, images, Word (`.doc` / `.docx`), plain text, CSV, Excel (`.xls` / `.xlsx`) |

Files can also be pasted or dragged onto the input.

> Attachments **persist across the conversation**. A file you upload on your first message is still in context ten turns later, and its extracted text is cached so it is never re-processed.

### Add Tool(s)

Pick from your connected MCP tools — see [MCP & Tools](./tools) — so the agent can call them while composing its answer. The picker is searchable and shows a check against each selected tool.

# Understanding Responses

A response renders as a single flowing message, not a set of tabs:

1. **Your prompt**, in a bubble
2. **The agent header**, with an expandable **"Thought for `{duration}`"** disclosure — click it to see the reasoning steps the agent took
3. **The answer** — text, and any widgets
4. **The toolbox**

## What an answer can contain

| Response type | What it looks like                                                                |
| ------------- | --------------------------------------------------------------------------------- |
| **Text**      | A written, markdown-formatted explanation                                         |
| **KPI**       | A headline metric card with an optional trend indicator                           |
| **Table**     | A paginated, scrollable data table                                                |
| **Chart**     | An interactive visualisation — bar, line, pie, scatter, and more                  |
| **File**      | An inline preview card for a PDF, CSV, or image the answer produced or referenced |

A single response can contain several widgets alongside explanatory text.

## Clarifying questions

The agent can pause mid-answer to ask you something — an ambiguous date range, which of two similarly-named tables you meant. Answer it, or **Skip**. The input is hidden while a question is pending.

## The message toolbox

| Action         | What it does                                                                                 |
| -------------- | -------------------------------------------------------------------------------------------- |
| **Copy**       | Copy the message content to the clipboard                                                    |
| **Share**      | Opens the [Share Message](#sharing-a-conversation-or-an-answer) dialog                       |
| **Confidence** | A colour-coded icon — hover for the overall confidence percentage and a per-widget breakdown |

The confidence icon only appears when the agent produced a confidence score for that answer.

Your own messages have their own hover toolbox with **Edit** and **Copy** — edit a prompt to re-ask it without retyping.

## Rating an answer

Thumbs up and down are **per widget**, not per message. Select a widget in the side panel and the **Like** / **Dislike** controls appear.

> These do real work. A thumbs up promotes the query behind that widget into the agent's [Verified Queries](./agents#verified-queries). A thumbs down feeds the agent's [Memory](./agents#memory).

## Platform help

You can also ask how to use Supaboard itself — *"how do I schedule a dashboard report?"* The agent reads the live product documentation and answers in text. No widget is produced for these.

# Pinning Widgets

Charts, tables, and KPIs generated by the agent can be saved elsewhere — and they stay live, re-querying your data whenever they're loaded.

1. After the agent responds, widgets appear in the **right-side panel**.
2. Check the ones you want to keep.
3. Click **Pin** (*"Pin widgets to dashboard"*).
4. Choose a destination:
   * **Dashboard** — an existing one, or create one inline
   * **Datasite** — send the widgets to a [datasite](./datasites), where the pinned query is used verbatim and only the presentation is rebuilt
5. Confirm.

# Sharing a Conversation or an Answer

Two entry points, one dialog:

* **The whole conversation** — the share button in the topbar
* **A single answer** — **Share** on any message

| Option              | Behaviour                                                                                             |
| ------------------- | ----------------------------------------------------------------------------------------------------- |
| **Invite by email** | Adds the person to the conversation and grants them Ask Analysts access if they don't already have it |
| **Member list**     | Everyone who currently has access                                                                     |
| **Public link**     | Expand the row and click **Copy Link** for a URL anyone can open                                      |

A link shared from a specific message opens the conversation scrolled to that message. Public views are read-only — no pinning, no replying.

> The dialog reports "Invite sent", but **no invitation email is currently delivered**. Send the person the link yourself.

# Managing Conversations

Every action is in the three-dot menu on the conversation's row in the list.

## Rename

Opens a modal with a text input. The default name is *"Untitled Conversation"* until renamed.

## Delete

Opens a confirmation. Deleting a conversation permanently removes it and all its messages. **This cannot be undone.**

## Control Access

> **Conversations are visible to your whole workspace by default.** A conversation with no access list can be opened by any workspace member. **Control Access restricts a conversation — it does not share one.**

Open **Control Access** and select people to limit the conversation to that list. Your own entry and the workspace Owner cannot be removed.

If you're discussing something sensitive, set Control Access at the start rather than after the fact.

# Permissions

## Who can open Ask Analysts

Access is controlled by the **Ask Analysts** feature flag on each user's workspace membership — see [Workspace — Feature Access](./workspace#feature-access).

## Reading versus asking

| Action                                     | Minimum role                 |
| ------------------------------------------ | ---------------------------- |
| Open Ask Analysts and read conversations   | Viewer (with feature access) |
| Send a message                             | Editor                       |
| Pin widgets to a dashboard or datasite     | Editor                       |
| Rename, delete, or restrict a conversation | Editor                       |

A Viewer with the Ask Analysts flag can read every unrestricted conversation but cannot ask anything.

## Who can use a specific agent

Each agent has its own **User Access** list. If an agent is restricted, only people on that list can select it. Unrestricted agents are available to everyone with Ask Analysts access.

# Quick Reference

| I want to…                                | How                                      |
| ----------------------------------------- | ---------------------------------------- |
| Ask a question                            | Type in the input bar and press Enter    |
| Get a more thorough answer                | Switch to **Deep Dive**                  |
| Get a faster answer                       | Switch to **Concise**                    |
| Dictate instead of typing                 | Click the microphone                     |
| Add a document for extra context          | **+** → **Add File(s)**                  |
| Let the agent call an external tool       | **+** → **Add Tool(s)**                  |
| See how an answer was reached             | Expand **Thought for …**                 |
| Check how confident the agent is          | Hover the confidence icon in the toolbox |
| Save a chart to a dashboard               | Check it in the side panel → **Pin**     |
| Share a specific answer                   | **Share** on the message                 |
| Restrict a conversation to certain people | Three-dot menu → **Control Access**      |
| Return to a past conversation             | Search for it in the conversations panel |
| Start fresh                               | **New Conversation** in the topbar       |