Creating an AI Analyst¶
An AI Analyst is a named persona that business users chat with in Explorer, and in Slack or Teams if you deploy it there. The data team usually builds AI Analysts in Studio. First they connect data sources and create models, metrics and glossary terms. Then they create the AI Analyst, give it instructions and data access, test it, and deploy it.
Availability
- Who can do this: members with the Edit capability in the active Workspace. Capabilities are granted per Workspace, for example through the Administrator or Editor preset.
- Plan: your plan sets how many AI Analysts your organization can have and how many models each one can use. On a self-managed deployment, your licence sets these limits.
- Deployment: Actian SaaS and self-managed deployments.
Before you start¶
- Connect at least one data source and build the models the AI Analyst should query. Add metrics and glossary terms where they help.
- Switch to the right Workspace. An AI Analyst belongs to the Workspace you create it in, and it can only use that Workspace's semantic layer. Members of other Workspaces never see it.
Create an AI Analyst¶
- In Studio, click AI Analysts in the sidebar.
- Click Create Analyst.
- Enter a Name (up to 120 characters), for example "Revenue Analyst".
- Click Create.
Studio opens the new AI Analyst's page.
Each name must be unique among the AI Analysts in the Workspace. If the name is taken, the dialog shows An analyst with this name already exists. The name of a deleted AI Analyst can be used again.
What a new AI Analyst can do¶
A new AI Analyst starts with no data access and isn't deployed.
| Setting | Starting value | Effect |
|---|---|---|
| Model access | Choose specific models, with no models selected | It can't query any data. |
| Glossary access | Only terms with these tags, with no tags selected | It can't see any glossary terms. |
| Deployment | Not deployed (Studio users only) | Business users can't see it in Explorer, Slack or Teams. |
Recommended setup order¶
- On the Overview tab, add a description and pick an avatar.
- Write Instructions that give the AI Analyst business context.
- On the Models tab, choose which models, dimensions and measures it can use. The Metrics tab then shows the metrics that this model access makes available.
- On the Glossary tab, choose which glossary terms it can see.
- Add Suggestions and Saved prompts for business users.
- Test it in the Test your analyst panel.
- Deploy it to Explorer, Slack or Teams.
The AI Analyst page¶
The page has a header, a row of tabs, and the Test your analyst panel on the right.
Header¶
| Control | What it does |
|---|---|
| Not deployed / Deployed | Opens the deployment menu (Slack, Teams, Explorer). The status can read Deployed while Slack or Teams channels are still added, even after you choose Studio users only. See Access Management. |
| Monitoring | Opens Tasks (every conversation with this AI Analyst) or Analytics (usage over time). See Monitoring. |
| Version history | Lists saved versions of the AI Analyst. Use View changes to compare a version with the current one, and Restore to go back to it. Restoring records a new version; the current version stays in the history. |
| Delete (trash icon) | Deletes the AI Analyst. See Delete an AI Analyst. |
A version captures the AI Analyst's name, description, avatar, instructions, model and glossary access, deployment state, Explorer audience and the list of specific users. Restoring a version also restores its deployment state, audience and list of users, so restoring a version that was deployed deploys the AI Analyst again. Restoring skips any model, dimension, measure, tag or user that has since been deleted. If the restored model access would exceed your plan's models-per-analyst limit, the restore is refused. Suggestions, saved prompts and Slack or Teams channels aren't part of a version.
Tabs¶
| Tab | Use it to |
|---|---|
| Overview | Edit the name, avatar and description, and check how complete the setup is. |
| Instructions | Write the guidance the AI Analyst follows on every question. See Instructions. |
| Models | Choose which models, dimensions and measures the AI Analyst can query. See Models. |
| Metrics | See the metrics available to the AI Analyst. Metrics follow model access: granting a model also grants the metrics it needs. |
| Glossary | Choose which glossary terms the AI Analyst can see: All terms, Only terms with these tags, or All terms except these tags. |
| Suggestions | Manage the suggestion cards business users see in Explorer. See Suggestions. |
| Saved prompts | Manage slash-command prompts for this AI Analyst. See Saved prompts. |
If the window is narrow, the last tabs move into the … menu at the end of the tab bar.
Changes on a tab aren't saved automatically. The bar at the bottom of the tab shows Unsaved changes until you click Save changes. If you switch tabs with unsaved changes, Studio asks Discard unsaved changes?: click Keep editing to stay, or Discard changes to leave without saving.
Saved changes take effect from the next question, including for an AI Analyst that is already deployed. Deploying doesn't freeze a copy of the configuration.
Overview tab¶
Below the name, avatar and description fields, the Overview tab summarizes the setup in three sections:
| Section | Rows |
|---|---|
| Foundation | Instructions configured, Models connected, Metrics assigned, Glossary terms |
| Distribution | Deployments (Slack and Teams channels), Suggestions, Saved prompts |
| Usage | Tasks run, Active users (30d) |
A configured row shows a green check and a count, for example "4 models". The Instructions configured row shows Ready instead. An empty row shows Not configured. The usage rows always show a count, and they leave out test runs from the Test your analyst panel. Hover over the ? next to a row for a short explanation.
If some tabs are greyed out¶
You can open an AI Analyst without the Edit capability if you hold Chat or Manage members in this Workspace and have Studio access, for example through the Workspace Manage members capability. If you do:
- Overview, Models, Metrics and Glossary are greyed out, and opening an AI Analyst takes you to Instructions.
- Instructions, Suggestions and Saved prompts are read-only.
- The deployment menu and Monitoring don't appear, and you can't delete the AI Analyst.
Ask a member who manages your Workspace to grant you Edit if you need to build AI Analysts. See Members.
Manage AI Analysts from the list¶
The AI Analysts page lists every AI Analyst in the Workspace.
- Card view / List view: switch between layouts. Your browser remembers the choice.
- Star / Unstar: starred AI Analysts appear first, most recently starred first. The rest are sorted by their last edit. Stars are personal: they only change your own list.
- Analyst actions menu on each AI Analyst: Open in Explorer, Duplicate or Delete.
Duplicate an AI Analyst¶
Open the Analyst actions menu on an AI Analyst and click Duplicate. The copy is named "
| Copied | Not copied |
|---|---|
| Description and instructions | Avatar (the copy gets a new one) |
| Model access mode, selected models and their dimension and measure choices, and model access tags | Explorer deployment, audience and list of specific users (the copy starts Not deployed) |
| Glossary access mode and tags | Slack and Teams channels |
| Suggestions and saved prompts | |
| Stars and version history |
A duplicate counts toward your plan's AI Analyst limit.
Delete an AI Analyst¶
- From the AI Analyst page: click the trash icon in the header, then click Delete to confirm.
- From the list: open the Analyst actions menu and click Delete.
Warning
Delete in the Analyst actions menu deletes the AI Analyst straight away, without asking you to confirm.
Deleting an AI Analyst removes it from Studio and Explorer, frees its name, and frees its place in your plan's AI Analyst limit. The deletion is recorded in the audit log. Studio has no option to restore a deleted AI Analyst. Treat deletion as permanent, and use Duplicate to keep a copy first if you might need it.
Plan limits¶
| Message | When you see it |
|---|---|
| Analyst limit reached | You create or duplicate an AI Analyst when your organization has used its AI Analyst allocation. The count includes AI Analysts in every Workspace of the organization. Deleted AI Analysts don't count. |
| Models-per-analyst limit reached | You save model access in Choose specific models mode with more models than your plan allows for one AI Analyst. The tag-based modes (Models with any of these tags and All models except these tags) aren't blocked by this limit. |
Both dialogs offer Upgrade, which opens Settings → Usage, and Contact us, which opens support chat. Contact us doesn't appear on air-gapped deployments.
Chart colors¶
Chart colors are no longer set for each AI Analyst. The organization's chart theme (series palette, typeface, and axis and grid colors) is under Settings → Charts. Anyone with Studio access can view it; saving changes needs the Manage organization capability.
Let the Steward help¶
The Steward can create an AI Analyst or edit an existing one for you, including its instructions, model and glossary access, and suggestions. It proposes the change and applies it only once the change is approved. The Steward can't deploy an AI Analyst or change who can use it: you always do that yourself in Studio.