AIStora turns CSV files into something you can question. Upload a spreadsheet export, check its quality, get an automatic profile of every column, and ask questions in plain English. All calculations run on AIStora's server; the AI model only ever sees your column names, types and your question, never the rows themselves.
Fastest way to try it
On the Databases screen click Try the sample dataset. It loads a small coffee-shop business (180 customers, 1,200 orders, 13 products) so you can explore every feature in two minutes with realistic data. You can also download the sample files (customers, orders, products) and a deliberately messy spreadsheet export to see what Auto clean does.
A database is a folder for related CSV files, for example one per client, project or month. Click New Database, give it a name and press Enter. You can rename or delete it later from the card.
Drop one or more .csv files on the upload area. Each file becomes a table named
after the file; use Rename to give it a friendlier name. What works:
.xlsx) are not accepted yet: save each sheet as CSV first.Limits on this server: files up to 5 MB, 8 tables per database and 3 databases per account.
Click Auto clean next to a table to profile it. AIStora shows what it
found and what it would change, and only acts after you approve. It never edits your
original table: it creates a new _cleaned copy next to it. Fixes it can apply:
Customer Name becomes Customer_Name) and name blank columns.N/A, null, none into empty cells.$1,234.50, USD 954, (20.00), 12%, 1234,50.Yes/No, Y/N, TRUE/FALSE to true/false.Missing values are reported, never guessed or filled in.
When a database has two or more tables, AIStora asks the AI to propose how they connect
(for example orders.customer_id to customers.customer_id) from the
column names and types alone. Detected relationships let you ask questions that span tables,
such as revenue per customer segment. Use Re-detect relationships after
uploading more files.
Open the chat screen and click EDA report. In about a second you get a deterministic profile of every table: missing values, duplicates, numeric distributions with outlier flags, category breakdowns, date ranges and correlations, plus plain-language findings. It runs entirely on the server, uses no AI, and can be downloaded as JSON.
Click Auto analyze for an instant dashboard of what stands out. AIStora
ranks your columns by name and type (a measure such as order_total, dimensions
such as status or channel, a date column, name-like columns), then
computes totals by category with shares, the month-by-month trend with growth, the top
products or customers and how concentrated they are, the distribution of the main measure,
missing-value flags, and — when tables link through an id — breakdowns across tables such as
revenue by customer segment. Every card has a chart, the numbers behind it, and an
Ask about this button that sends the follow-up question to the chat. It uses no AI,
so it is free and instant.
Type a question or click one of the suggested chips. The AI plans a sequence of safe operations (filter, group, join, rank, bucket by month) and AIStora executes them locally, then shows the result table and the steps it took in the Agent activity panel. Questions that work well:
Use the column names you see in the schema panel; the AI cannot guess a column that does not exist. Ask for a "chart" to get a bar or line chart; the chart is rendered by an external service, so AIStora asks your permission before sending it the aggregate numbers.
AI questions are limited to 10 per account per day during the beta. EDA reports, cleaning and uploads are unlimited.
Within a database, the AI remembers the recent conversation so follow-up questions work ("now only for 2025"). Clear memory forgets that conversation. Thumbs up/down on an answer teaches AIStora which plans worked; Clear learning deletes those saved evaluations for the database.
AIStora is in public beta. Found a problem? Email jamiejwei@gmail.com.