The AI spreadsheet,
supercharged for every row.
Cube is the AI spreadsheet for every row of your data. Type a prompt, run it across every row, export the CSV. It's that simple. If you can use a spreadsheet, you can use Cube.
No new tool to learn. It's just an AI spreadsheet, the same rows and tables you already know. Describe the task in plain language, and Cube runs it across every row.
Your rows go in plain. They come out done.
No formulas to write, no charts to build, no pivot tables or dashboards to set up. You describe the result in plain language, and the AI does the data analysis row by row, spotting trends and turning a raw column into insights you can use.
| product | price |
|---|---|
| Bamboo cutting board | $24 |
| Wireless earbuds | $59 |
| Cast iron skillet | $38 |
| LED desk lamp | $29 |
| product | description | category |
|---|---|---|
| Bamboo cutting board | Eco-friendly, knife-friendly. | Kitchen |
| Wireless earbuds | Deep bass, all-day battery. | Audio |
| Cast iron skillet | Pre-seasoned, built to last. | Kitchen |
| LED desk lamp | Warm and cool light modes. | Lighting |
Whatever your list is. Cube runs it row by row.
An AI spreadsheet handles any repetitive per-row task you'd do by hand, across any industry. The tasks below all start the same way: ask in plain language, run it on the whole sheet. Writing, classifying, extracting, and per-row analysis across your sheets, all from one prompt.
Write product descriptions
Turn a bare product name into copy that's ready to publish, for your entire catalog at once.
Generate meta descriptions
Give Cube a list of URLs and get a clean, length-appropriate meta description for each page.
Parse resumes
Pull structured facts out of free-text resumes: years of experience, skills, last title, location.
Tag and triage messages
Classify every ticket, review, or message by type and sentiment so the right ones rise to the top.
Summarize at scale
Condense hundreds of survey responses, reviews, or notes into a one-line theme per row.
Structure listing text
Extract beds, baths, price, and features from messy listing descriptions into tidy columns.
Fix messy fields
Standardize addresses, split names, pull out ZIPs or phone numbers. The universal spreadsheet chore.
Enrich and classify leads
The go-to-market staples: classify companies by industry, score against your ICP, or call an enrichment API per row.
Translate at scale
Translate product copy or support replies, or any text column into one language or ten, row by row.
Plus anything with a REST API: your CRM, internal tools, vendor APIs, your own backend. If it speaks HTTP, the HTTP column can run it on every row.
A 2000-row sheet shouldn't take a week.
You're pasting answers from a chatbot, one row at a time.
You ask the AI tools to generate one result, then tab over, paste, copy, tab back. 1999 more rows to go. Your wrist hurts. You've made three mistakes.
The tool that does this costs hundreds a month. And there's a waitlist.
You need to run one job, this week, for work that's already late. You need a tool, not a platform.
Your automation workflow broke again.
The HTTP module timed out. The loop forgot row 47. You're debugging YAML at 11pm. There's a better way.
Three steps. No tutorial required.
Upload your CSV.
Export your data from Google Sheets or Microsoft Excel as a CSV, then drop the file in. Or start blank and paste rows. CSV in, CSV out. No row caps disguised as "credits."
products.csv → imported
→ 1,847 rows loaded
Type what you want.
Create a new AI column in plain language. Reference other columns with /column, then pick a model. No formulas to learn.
Describe /product
in one line.
Click run. Get coffee.
Cube runs your AI prompt on every row in parallel to generate the results, whether you are cleaning fields, classifying rows, or analyzing text in the table. Watch the column fill in, then export the finished file back to CSV.
1,847 / 1,847 rows ✓
→ Export
Pick a prompt. Watch it run.
A preview of how an AI column fills in. Pick a prompt below and hit run.
What's the sentiment of /message? One word.
| message | sentiment |
|---|---|
| App keeps crashing on login | |
| Love the new dashboard! | |
| How do I export my data? | |
| Charged twice this month | |
| Can you add dark mode? |
This is a preview with sample data. In Cube you bring your own rows, prompt, and model.
Run it on your own listTwo columns. You bring the data sources.
Cube is the engine that runs your prompts and API calls across every row. Two simple column types let you build finished work from flat tables: writing, classifying, extracting, and per-row analysis, all based on your own data. The AI models, the data providers, the API keys: all yours. No middleman. No vendor lock-in.
AI Column
Connect your OpenRouter key, pick from 300+ models, write one prompt, run it on every row. Bulk AI prompts without the copy-paste. Fast and cheap or smart and slow, your call. Toggle web search and web fetch when you need them.
HTTP API Column
Plug in any REST API you already use (Clearbit, Hunter, Apollo, your CRM, your own internal API, anything with an endpoint). Define the request once with /column and /api_key substitution. Cube runs it per row and extracts JSON fields via JSONPath into new columns.
Web search + fetch
Optional toggles on any AI column. Give the model the live web when the answer isn't in its training data. Off by default. You only pay when you flip it on.
Your keys. Your bills.
Cube doesn't resell AI tokens or proxy API calls. You hold the keys. You pay your providers directly. No markup, no credits, no surprise invoices. We're invisible on your bill.
Rows flow in. Agents pitch in.
Cube doesn't stop at CSV. Every sheet has a webhook URL other tools can send rows to, and the whole spreadsheet speaks MCP so AI agents can work it for you.
Incoming webhooks
Every sheet gets its own webhook URL. Point a form, Zapier, Make, your CRM, or your own backend at it. Each JSON payload lands as a new row, live in your open tab, with a delivery log and field mapping in the app.
POST /api/webhooks/… {"email": "ada@example.com"}
→ new row appended · live in your open sheet
MCP for AI agents
Connect Claude, Codex, or Cursor and work your sheets by chatting. Agents create tables, import CSVs, edit rows, and run enrichments through 19 tools, gated by scoped access tokens. Set it up in under a minute.
"Import this CSV and classify every company by industry."
→ table created · AI column runs on 1,847 rows
Pay your model provider, not us.
No seats. No trials. No "talk to sales." Get the full AI spreadsheet on the free plan, bring your own API key, pay your provider for what you use.
- › 2 tables per account
- › 10,000 rows per sheet
- › 80 columns per sheet
- › Bring your own OpenRouter API key
Questions, answered.
What does the free plan include?
2 tables per account, 10,000 rows per sheet, and 80 columns per sheet. That's the whole product: AI columns, HTTP API columns, webhooks, and MCP access. You bring your own OpenRouter API key and pay your provider directly. Cube adds no markup.
Do I need my own API key?
Yes. Connect an OpenRouter key once and every AI column can use any of 300+ models. HTTP API columns use the keys of whatever service you call. Your keys stay yours, and your usage is billed by your providers, not by us.
How is this different from pasting rows into a chatbot?
A chatbot answers one message at a time. Cube runs your prompt on every row in parallel and writes each answer into the right cell. A 2,000-row job is one prompt and one click, and you export the finished CSV when it's done.
Can Cube call APIs that aren't AI?
Yes. The HTTP API column calls any REST API per row: enrichment providers, your CRM, internal services, anything with an endpoint. Define the request once with /column placeholders, and Cube extracts the JSON fields you pick into new columns.
How do rows get into Cube?
Import a CSV, paste rows, or start blank. Sheets also accept rows over webhooks: give any tool the sheet's webhook URL and every JSON payload it sends becomes a new row, live in your open tab. Export back to CSV whenever you want.
Can AI agents use Cube?
Yes. Cube has a built-in MCP server, so Claude, Codex, Cursor, and other MCP clients can create tables, import data, edit rows, and run enrichments for you. See the MCP page for one-minute setup.
Stop pasting. Start running.
Whatever the task, your next 2000-row file is one prompt away.
Start free