Sessions
A Session is a conversation with the AI. You ask questions in natural language, the AI queries your data through the semantic layer, and returns answers, charts, and insights — all within a single, persistent conversation thread.
Sessions are available to all users.
Starting a Session
- In the workspace, click Sessions (or the + button to start a new one).
- A new, empty session opens.
- Select a Custom Model from the model selector in the input bar. The model determines which data and semantic layer the AI will use to answer your questions.
If a default model has been configured by your administrator, it is pre-selected automatically.
The Input Area
The input bar at the bottom of the session is the main interaction point.
Toolbar Buttons
| Button | Function |
|---|---|
| Paperclip | Open the file picker to attach tabular files (CSV, Excel, etc.) |
| Model Selector | Choose or switch the Custom Model for this session |
| Microphone | Start voice recording — your speech is transcribed and inserted into the input field |
| Submit | Send the message (also triggered by Enter) |
| Stop | Interrupt a running AI response |
Keyboard shortcuts in the input field:
| Shortcut | Action |
|---|---|
Enter | Submit the message |
Shift + Enter | Insert a new line |
Arrow Up (empty input) | Edit the last user message |
Ctrl/Cmd + R (empty input) | Regenerate the last AI response |
Input Shortcuts
Typing a specific character at the start of the input activates a picker:
| Character | Opens |
|---|---|
/ | Prompt commands — your saved prompt shortcuts |
@ | Chart selector — select existing charts to reference or combine into a dashboard |
Attaching Files
Click the paperclip icon or drag and drop files directly into the session window.
Supported File Types
Lakehousecat is focused on structured, tabular data. Only tabular file formats are supported for upload:
| Format | Examples |
|---|---|
| CSV | Comma-separated values |
| Excel | .xlsx, .xls |
Tabular files and sessions: Each tabular file you upload creates a temporary data source scoped to the current session. It is available for analysis within this session only and does not appear in the Data section.
Voice Input
Click the microphone icon in the input bar to start recording. Speak your question — the audio is transcribed and inserted into the input field. You can then review and edit the text before sending.
After transcription, a magic icon (✨) appears next to the transcribed text. Click it to automatically optimize the phrasing — the spoken input is rewritten into a clean, well-structured query. This is particularly useful when dictating longer descriptions or system prompts.
Behavior settings:
| Setting | Description |
|---|---|
| Auto-Send | Automatically submits the message after transcription completes, without manual confirmation |
| Auto-Optimize | Automatically applies the magic optimization after transcription. Can be disabled in Settings → Behavior if you prefer to review the raw transcription first |
Voice input is especially useful for entering model descriptions and system prompts. Dictate your intent naturally — then use the magic icon to clean it up.
Using Saved Prompts
Type / in the input field to open the prompt command picker. A list of your saved prompt shortcuts appears. Continue typing to filter by name. Select a prompt to insert its full text into the input field.
You can then review or edit the inserted text before sending.
For creating and managing saved prompts, see Prompts.
Selecting Charts for Dashboards
Type @ in the input field to open the chart selector. Select two or more existing charts to attach them to your message. When you submit with multiple charts attached, the AI automatically generates a dashboard combining the selected charts.
Selected charts appear as preview tiles above the input field. Remove a chart by clicking the × button on its tile.
Session History
Navigate to Workspace → Sessions to access all your sessions.
Sections
| Section | Description |
|---|---|
| My Sessions | All sessions you have created |
| Shared by Me | Sessions you have shared with others |
| Shared with Me | Sessions others have shared with you |
List Features
| Feature | Description |
|---|---|
| Search | Filter sessions by title |
| Tag Filter | Filter by assigned color tag |
| List / Cards View | Toggle between display modes |
Session Actions
Hover over a session entry to access the available actions.
In My Sessions
| Action | Description |
|---|---|
| Open | Click the session to continue the conversation |
| Rename | Click the pencil icon to edit the session title inline |
| Clone | Create an independent copy of the session |
| Share | Open the sharing dialog |
| More (⋯) → Tag | Assign a color tag to the session |
| More (⋯) → Delete | Permanently remove the session |
In Shared by Me
| Action | Description |
|---|---|
| Unshare | Revoke sharing for all recipients at once |
In Shared with Me
| Action | Description |
|---|---|
| Clone | Create your own copy of a shared session |
Sharing a Session
- In My Sessions, hover over the session.
- Click More (⋯) → Share.
- In the sharing dialog, switch between Groups and Users tabs.
- Select groups or individual users to share with.
- Assign a permission level for each recipient:
| Permission | Description |
|---|---|
| Read | Recipient can view the session and its messages |
| Write | Recipient can continue the conversation |
| Share | Recipient can share the session further |
- Confirm to apply.
The session appears in the recipients' Shared with Me section.
To stop sharing, go to Shared by Me and click More (⋯) → Unshare.
For a full overview of sharing across all object types, see Sharing.
Session Persistence and Custom Model Dependency
How long are sessions stored?
Sessions are permanently stored in the Lakehousecat database. There is no automatic expiry or session limit per user — the storage backend uses a PostgreSQL integer identifier, which supports approximately two billion sessions before overflow. In practice, sessions persist indefinitely until explicitly deleted by the user.
Dependency on Custom Models
Sessions and the charts they generate depend on the Custom Model that was active when the session was created. This dependency is deeper than it may appear:
When a Custom Model is trained, it generates Star Views in ClickHouse — denormalized analytical structures that are the actual data source for all charts and queries in that session. If the Custom Model is deleted, these ClickHouse views are also deleted, breaking the entire view chain. Any chart that was generated using that model will no longer resolve and will return errors.
The same problem occurs when Custom Model sharing is revoked: a user who loses access to a Custom Model also loses the ability to render any chart in their sessions that was generated through that model.
Deleting a Custom Model in active use will break all charts and sessions that depend on it — for all users who have access to those sessions. This cannot be undone without re-training the model.
Recommended approach:
- Keep productive Custom Models active and shared — do not delete them.
- For iterative development, create a new Custom Model for each significant change. This preserves the existing model and all charts that depend on it.
- Retire old models only after confirming that no active sessions or charts reference them.
Tips
- Rename sessions: Sessions are auto-titled from the first message. Rename them immediately after starting a focused analysis so they are easy to find later.
- Use prompts for repeating queries: Save frequently used queries as prompt commands and invoke them with
/for faster access. - Scope tabular uploads: Tabular files are scoped to the session — upload them fresh in each new session where you need them.
- Clone before experimenting: If you want to try a different angle on an existing conversation, clone the session first to preserve the original.