Skip to main content
Larkup understands structured data. Upload an Excel or CSV file and the AI can aggregate rows, compute totals, compare categories, and render the result as a bar chart, line chart, or table inside the Chat conversation.

​
Loading the data

Go to Data → Files and drag in your .xlsx, .csv, or .json file. Larkup detects the structure automatically. Column types, row count, and the data schema are shown before you confirm. Workbooks with multiple non-empty worksheets are loaded as separate named datasets, which lets you ask about one sheet or compare them in the same chat.
Excel file staged in the Files tab with column types detected

A 500-row Excel file staged for upload with column schema preview

Click Start Indexing to begin. Progress is shown in real time.
Indexing progress for Excel upload

Indexing progress for a structured data file

​
Asking analytical questions

Once indexed, open Chat and ask anything about the data:
“Show revenue by region as a bar chart” “Which sales rep had the highest net revenue in Q1?” “Give me a monthly breakdown of gross sales”
Larkup reads the relevant rows, computes the answer, and renders the chart or table inline. Cartesian charts show names for both axes and include units when the available data supplies them.
Bar chart rendered inline in Chat

A bar chart generated in Chat from a sales dataset

Detailed chart answer with analysis and later suggestions

Chart with a later summary and suggestions

Each chart or table response includes Copy and CSV actions so you can take the visible result into another tool. You can also ask Chat to export the current answer as an Excel, CSV, or PDF file—for example, “Export this result as Excel.” Larkup creates a real download from the result already shown; it does not run a new, broader database query for a direct export follow-up. Later questions build on the previous answer. You can drill down into a specific category, change the chart type, export the exact result, or ask for a written summary without uploading the data again.

​
Feedback and exact repeats

The Like and Dislike buttons are saved with the conversation and remain selected after you reload it. Liking a plain-document grounded answer also saves it in the project’s temporary answer cache, so the same exact question can reuse the answer without another search—even in a new chat—while the available sources and Assistant group settings are unchanged. Unrated consecutive exact repeats can also reuse their immediately preceding grounded result. Disliking removes the rejected answer from reuse and records negative feedback, so the next exact repeat performs a fresh search. Unliking removes that feedback record. Changing, moving, adding, deleting, selecting, or deselecting source data also changes the source scope and forces fresh retrieval. The saved thumb selection belongs to chat history, while Settings > Larkup cache globally reports the liked and disliked answer-feedback records across all projects. Clear cache removes both kinds of temporary feedback records without deleting chat history. For questions that relate worksheets or files, ask for the comparison explicitly—for example, “Join Orders to Returns by Order ID and show returned-order profit by region.” Larkup can stage the selected sheets together for code-based analysis while keeping the raw rows out of the conversation context.