Multimodal AI work automation
Repetitive work ends here.
Tell NextRows what needs to get done, in plain words. We're building multimodal AI that reads the screen, parses documents, and completes the whole process on your schedule.
Every morning at 8, order minced garlic from supplier mall A and log it in the ingredient sheet.
- Open supplier mall A
- Find minced garlic 1kg
- Place the order
- Log to ingredient sheet
The cost of repetitive work
How it works
Describe the task
Type what you need in natural language: “Every morning at 8, order minced garlic from our supplier mall and log it in the ingredient sheet.”
AI plans the work
A multimodal model is being developed to read the screen and your documents, breaking the goal into concrete actions.
It runs on schedule
Workflows are being built to run on your schedule and to re-plan when websites or tools change.
Why NextRows
Traditional automation
- Months of implementation and setup
- Dedicated external engineers required
- Breaks when processes or websites change
NextRows
- Starts with a sentence, not a spec document
- No engineers. Just describe the work
- Built to adapt when processes change
The technology behind it
Typed, documented API
A Hono API with OpenAPI-described routes and Zod request schemas.
Real browser rendering
Pages load in headless Chromium through puppeteer-core, with animations paused and hidden elements removed before content is captured.
Pages, PDFs and text as input
HTML goes through Readability-based cleanup, PDFs are converted to Markdown, and plain text is accepted as is. One extract request takes up to 20 sources.
Schema-guided output
Extraction runs through a model provider layer built on the Vercel AI SDK. Pass a JSON schema for predictable output, or let the system generate one.
- 01Input
- Web page URLs
- PDF documents
- Raw text
Up to 20 sources per request
- 02Render
- Headless Chromium
- puppeteer-core
- PDF to Markdown
Raw text skips this step
- 03Normalize
- Readability cleanup
- HTML to Markdown
- Source text merged
Clean text for the model
- 04Extract
- Google Gemini models
- Vercel AI SDK
- JSON schema
Schema-guided output
- 05Deliver
- Table data
- JSON
- OpenAPI routes
Run, table and JSON endpoints
Extract request body
{
"type": "url",
"data": ["https://fortune.com/ranking/global500"],
"prompt": "Extract the top 10 companies from the list in a table format."
}Enforced by the request schema
- typeurl | text
- data1 to 20 sources
- promptoptional, up to 2,000 characters
- schemaoptional JSON schema
- Web app
- REST API
- Browser extension
- Desktop app (Tauri)
What you can automate
Document processing
Parse invoices, forms, and spreadsheets. Enter the data where it belongs.
Scheduled operations
Recurring orders, reports, and updates. On time, every time.
Data collection
Gather information across websites and keep it up to date.
Everyday office work
Fill sheets, update systems, and move data between tools.
How teams start
Book a call
Tell us about the repetitive work you want off your plate.
Scope it together
We map your workflows and find where automation saves the most time.
Explore a pilot together
Discuss where a pilot could fit and what it should prove first.
A marketplace for automation
We're building a place to share workflow templates and discover automations from other teams, so you can start from a template instead of a blank page.
Frequently asked questions
Is NextRows available today?
NextRows is in active development with early partners. Book a call and we will walk you through the current state and explore whether a pilot fits your team.
What kinds of work can it automate?
Repetitive computer work: document processing, scheduled operations, data collection, and everyday office tasks like filling sheets and moving data between tools.
Do we need engineers to use it?
No. You describe the work in plain words. The AI is being built to plan and run the process for you.
How is our data handled?
Security-sensitive deployment is a core design goal: quantized on-device models and a local RAG pipeline are in development for exactly these environments. We are happy to discuss the details on a call.
Which languages does it support?
English, Japanese, Korean, and Chinese.
What does it cost?
Pricing is not final yet. Early partners help shape it - book a call to talk about your case.