AI Assistant¶
The AI Assistant is an interactive chat embedded in the Report Editor that can inspect your report, create and modify elements, manage data sources, parameters, and dictionaries, and answer questions -- all through natural language.
To use the AI Assistant, your role must have the AI permission (see Roles) and at least one AI Integration must be configured in the workspace. If no integration is configured yet, clicking New Chat offers to take you to the AI Integrations settings page (if your role can manage integrations) or asks you to contact your administrator.
Conversations¶
Each conversation is tied to the report you were editing when it was created. The model dropdown at the top of the chat lets you pick which AI Integration to use. Each conversation remembers its model choice, so you can switch models mid-conversation without losing history. New conversations default to the model from your most recent chat.
What the Assistant Can Do¶
Every time you send a message, the assistant automatically receives an overview of your current report -- pages, elements, data sources, parameters, dictionaries, and custom components -- so it can respond in context without you having to explain your setup.
Beyond answering questions, the assistant can take actions on your behalf. These actions appear in the chat as collapsible detail rows so you can review what happened. The assistant can:
- Inspect any page structure, data source, parameter, dictionary, custom component, theme, or current runtime data
- Add, update, or remove elements on a page
- Add pages and navigate between them
- Create, update, and delete data sources and parameters
- Create dictionaries (report-scoped or global) and manage translation entries
- Browse workspace languages, external databases, and the file system
- Load reference documentation on demand for any component type or topic (data sources, text formats, ECharts, etc.)
The assistant can also leverage AI Skills -- user-defined instructions that are automatically included based on their scope, giving the assistant domain-specific context for the report you are working on.
Attaching Files¶
You can attach a file to a message and ask the assistant to work from it -- paste a screenshot of an existing dashboard and ask it to recreate the layout, hand it a PDF report to rebuild, or give it a CSV to turn into a data source.
To attach a file, click the paperclip button next to the message box, drag it onto the message box, or paste it directly into the chat with Ctrl+V. Attachments appear above the message box and can be removed before you send. You can attach files while the assistant is still answering, but you can send the next message only once it has finished.
You can attach PNG, JPEG, WebP, PDF, JSON and CSV files, up to 3 per message:
| File type | Limit per file |
|---|---|
| Image (PNG, JPEG, WebP) | 7 MB |
| 32 MB, 100 pages | |
| JSON, CSV | 256 KB |
Images and JSON/CSV files in one chat can add up to 20 MB in total. Once that is reached, you are asked to start a new chat.
JSON and CSV file names cannot contain { or }. Rename the file before attaching it.
Images and PDFs¶
Images are sent as attached. With Anthropic, very large images are scaled down to 2000px on the longest side.
PDFs are uploaded to your AI provider, which keeps them for a limited time that depends on the provider. A PDF works only with the integration it was uploaded with. If a PDF is no longer available, a message asks you to attach it again.
JSON and CSV¶
The assistant reads the full text of JSON and CSV files, so their content counts toward the model's context. When it creates a data source from an attached file, the data source gets the file's exact content. If the file is no longer attached, the assistant tells you instead of creating a data source with missing rows.
What you can attach depends on the AI integration
| Integration | Images | JSON / CSV | |
|---|---|---|---|
| OpenAI | yes | yes | yes |
| yes | yes | yes | |
| Anthropic | yes | yes | yes |
| DeepSeek | yes | no | yes |
The paperclip button offers only the formats the selected integration accepts, and explains why if you paste something it cannot take. For images, the model itself must also be able to read images; otherwise the request fails.
Reopening a Chat¶
A reopened chat shows earlier attachments as labels. Images and JSON/CSV files need to be attached again, also after switching integration. PDFs can be reused until the provider removes them. Removing a PDF before sending it, or deleting the chat, asks the provider to delete the file.
Tips¶
- Select the right page first. The assistant sees the currently selected page. Navigate to the page you want to work on before sending your message.
- Select an element for targeted changes. Select an element in the editor so the assistant knows which one you mean.
- Be specific. Instead of "make it look better," try "change the heading font size to 18px and set the color to dark blue."
- Use Skills for repeated context. If you keep explaining the same business rules or database structure, create an AI Skill instead.
- One page at a time. The assistant works best focused on a single page. For multi-page changes, work through them sequentially.