Appearance
AI profile
An AI profile is a setting that bundles the connection method, permissions, and fixed context for the LLM the AI agent uses. It's also used by LLM actions in workflows.
Create and manage AI profiles from global nav > Workspace settings > AI profiles. While chatting with the AI agent, you can switch the profile in use from AI profile in the input field.
Types
From New AI Profile, choose one of the following.
| Type | Description |
|---|---|
| Managed | Uses an AI model managed by Codatum. No API key is required |
| Custom | Uses an API key for OpenAI / Anthropic / Gemini / Azure / Vertex AI / OpenRouter / Fireworks, and more |
Custom might not be available, depending on your contract.
Creating and managing
- Open global nav > Workspace settings > AI profiles.
- From New AI Profile, choose Managed or Custom.
- Enter each item, then select Create.
You can open an existing profile from the list and change it with Update.
Settings
Name
The display name of the profile.
Model selection / provider / model
- Managed: choose a model tier (a selection based on intended use) under Model selection. You can't specify a provider or an individual model name.
- Custom: configure the provider, model, and API key, among other settings.
Context
Specify content to always reference when the AI agent runs (optional).
| Item | Description |
|---|---|
| Notebooks and folders | Loads a linked notebook, or the notebooks within a linked folder. You can also specify folders or notebooks within a teamspace |
| Plain text | Additional instructions or descriptions passed as text |
If you link a folder, the notebooks within it are loaded. If you link a folder within a teamspace, using that AI profile requires view permission on the target folder.
WARNING
When a workflow runs, the AI profile's context isn't loaded.
Tool permissions
Restricts what the AI agent can use, on top of the executing user's permissions.
| Item | Choices / content |
|---|---|
| Notebook access | All accessible notebooks / Current notebook only |
| Connection access | All accessible connections / Allow only selected / Deny selected |
| Connector access | Off / Allow only selected connectors / All connectors |
| Web search | Off / Allow specific domains / On |
| Web fetch | Off / Allow specific domains / On |
For domain specification, you can use formats such as example.com or *.example.com.
Sharing
Manage which members and groups can use the profile.
| Target | Description |
|---|---|
| Workspace Owner | Always included (fixed) |
| Workspace members | Grants access to all members |
| Group / individual members | Add from Add group or member. Choose the role: Owner or User |
- Owner: can edit the profile, among other things.
- User: can chat using this AI profile.
Save conversation message content to logs
Available for managed profiles. When turned on, the content of conversation messages is saved to logs.
Start screen
You can configure the welcome text shown when a chat starts, and suggested prompts (optional).
Default behavior
Shown when the corresponding feature is enabled for the workspace. Lets you choose which of the standard behavior instructions to disable. You can expand each item on screen to check its instruction content.
Turning an item off means that instruction isn't passed to the AI agent. Items are shown in the following groups.
Common
| Item | Content |
|---|---|
| Analyst persona | The basic stance for supporting data analysis, charts, and report creation |
| Response style | How to write professional, objective responses |
| Result interpretation | Summarizing the key points of execution results and confirming they match the request |
| Data analyst workflow | The steps for checking and breaking down work when proceeding with SQL and analysis |
| Problem-solving integrity | Checks to keep the basis, scope, and unit of reasoning consistent |
| Plan workflow | Using Plan for complex, ambiguous, or high-risk requests |
| Completion report | How to summarize the final response when work is complete |
Agent mode
| Item | Content |
|---|---|
| Agent mode autonomy | The policy of proceeding with analysis and edits without waiting for approval |
| Agent mode self-review | Checking and fixing results after creating SQL or a chart |
There are currently no items for Ask mode. The items and content shown may change without notice.