Airfront
Airfront · Writing · Marketing
Airfront is an agentic Excel parser that turns messy spreadsheets into typed JSON for AI agents and data pipelines, reading table structure and cell formatting instead of guessing at a grid.
About Airfront
What Is Airfront
Airfront is a parsing service for one of the least glamorous problems in AI work: getting clean data out of real Excel files. The company calls what it does agentic Excel parsing, which is a fair description once you see the pipeline. Most spreadsheet tools assume a tidy grid with a header row on top. Business files rarely look like that. A price list has three tables stacked on one sheet, a merged banner across row one, and a note in cell F42 that explains what the asterisks mean.
The company's pitch is that spreadsheets aren't structured data at all, so ordinary spreadsheet parsing starts from a wrong assumption. They're a visual format that humans read by position, color, and layout. Airfront treats a workbook the way a person does: find the tables first, work out where each one starts and ends, rebuild the headers, then read the values. It also keeps the formatting as data, so a red cell or a bold total arrives as a flag instead of disappearing into plain text.
Output is typed JSON that matches a schema you provide. That's the part that matters for an AI data pipeline. An agent that receives {"unit_price": 12.5, "currency": "EUR"} can do arithmetic. An agent that receives a pasted text dump of the same sheet spends its context budget guessing which column held the price. Airfront is aimed at teams building agents, RAG systems, and internal data tools that keep hitting that wall.
The current state is early. Airfront runs on an early access list, requests go through email, and there's no public pricing page or documentation site. The homepage shows a command line example and benchmark numbers, which tells you the first users are expected to be comfortable in a terminal.
Getting Started
- Request early access by emailing the team from the address listed on the site, and include your name, company, and the kind of files you want to parse.
- Get set up with the
airfrontcommand line tool once access comes through, then confirm your account works on a single workbook. - Write a small schema file for the data you need, such as pricing rows with part number, description, unit price, and currency.
- Run
airfront parse "Price List.xlsx" --schema pricing > data.jsonand check the JSON against the source sheet before you wire it into a pipeline.
Product Information
A quick look at Airfront's pricing, supported platforms, and performance.
Best for
The users, tasks, and scenarios where this tool fits best.
Users
- AI engineers building agents that read business files
- Data teams with a folder of inherited Excel workbooks
- Analysts who hand data to a model
Tasks
- Extracting pricing tables from supplier sheets
- Turning messy spreadsheet data into rows for a database
- Feeding spreadsheet content to a RAG system
Scenarios
- A quarterly reporting job that depends on twelve workbooks from twelve different teams
- Onboarding a new data source where the file layout is unknown
- Agents that need a number, not a paragraph
Key features
Table Detection Across a Messy Sheet
Airfront segments a worksheet into distinct tables and blocks before it reads anything. It handles tables that sit side by side, stacked vertically, or separated only by a blank row, and it sets titles, notes, and stray labels aside instead of folding them into your rows.
Header Reconstruction
Single-row headers are the easy case. Airfront also rebuilds multi-row headers, merged cells, and repeated label groups, which is where plain extraction usually produces a column named Unnamed: 4. If a header is missing, it infers one from the surrounding labels and the shape of the data below.
Formatting Read as Data
Color, bold text, highlights, and number formats carry meaning in business files. A red row can mean discontinued. A bold row can mean a subtotal. Airfront captures those signals as structured flags, so downstream code can filter on them instead of losing the information the moment the file becomes plain text.
Formula Resolution and Type Normalization
Cells that hold formulas get resolved to their values. Dates, currencies, and percentages are converted to real types during the parse, so a price arrives as a number with a currency code rather than the string €12.50. That's the difference between a field you can sum and a field you have to clean again.
Schema-Validated JSON With Provenance
You pass a schema on the command line and get back JSON that conforms to it. Every value keeps provenance, which answers the question support teams ask most often: where did this number come from. The output drops straight into an agent, a RAG index, or a warehouse load without a translation layer.
Online-Judge Style Validation
Each parse is graded against expected outputs, the way a coding benchmark checks a submission. In practice that means the service reports whether it read the sheet correctly rather than returning confident nonsense. For a pipeline that runs unattended, a signal like that is worth more than a slightly higher accuracy score.
Pros and cons
Pros
- Reads structure before values, which is the actual reason most spreadsheet data extraction fails on real files.
- Keeps formatting as flags, so color coding and bold subtotals survive into your data.
- Schema-first output means your pipeline gets the fields it asked for, not a best guess at the column names.
- Benchmarked openly on SpreadsheetBench, including the tasks it still gets wrong.
Cons
- Early access only, with pricing and documentation not published, so you can't evaluate cost or read the API reference before you commit.
- The documented workflow is a command line tool, which puts it out of reach for non-technical teams that would otherwise benefit.
- It reads spreadsheets and doesn't write them back, so anyone expecting an Excel automation tool will need something else for the editing side.
- The headline benchmark number covers simpler tasks, and the company itself points out that a meaningful share of harder tasks still fail.
Frequently asked questions
It handles the Excel to JSON step: it locates tables, rebuilds headers, and returns typed records. It rebuilds headers, reads formatting cues, and returns data that matches a schema you supply.
Related content
Explore related tools, skills, and articles for Airfront.
Airfront Alternatives
Google EmbeddingGemma 2
Google DeepMind · Writing · Image · CodingGoogle EmbeddingGemma 2 is a 740M-parameter open model that turns text, code, images, audio, and video frames into vectors in one 768-dimension space, running on phones and laptops.
MePersona
MePersona, Inc. · Writing · Productivity · MarketingMePersona is an AI unified inbox for creators. It learns your voice, drafts replies to fan messages from 12 platforms, and sends only what you approve.
Fluence
Yay Software (Yannick Huchard) · Writing · Voice & Language · ProductivityFluence is an Android app that reads PDFs, EPUBs, Word files and web articles aloud, follows each sentence as it speaks, and answers questions about the text on the phone itself.