Data Intelligence

The problem with data today

You have files — CSV exports, spreadsheets, reports from your tools. When you hand one of these to an AI and ask it to analyze them, it does something surprisingly clunky: it reads the entire file as text. Thousands of rows of raw numbers, all at once.

This is slow. It uses a lot of tokens, which means it costs more. And the analysis is often unreliable, because the AI is trying to hold an enormous amount of raw data in its head all at the same time.

There had to be a better way.

What Synap does differently

Starting with version 1.0.5, Synap ships with a built-in data engine. When you ask Claude a question about a data file, it doesn't read the whole thing. Instead, it writes a precise query and gets back just the answer it needs.

Think of it as the difference between reading an entire book cover to cover to find one fact versus looking it up in the index. Same result, a fraction of the work.

This engine runs automatically, in the background, every time Claude touches a data file. You don't configure anything. You don't even know it's there. You just ask a question and get an answer.

What this looks like for you

Drop a CSV file anywhere in your mind. Then ask Claude something about it.

• "What were my top 5 products by revenue last quarter?"
• "Show me the monthly trend of customer acquisition cost."
• "Compare sales across regions — which is growing fastest?"
• "Join my customer list with my orders file and show me who hasn't ordered in 90 days."

You're having a conversation with your data. No formulas, no pivot tables, no exporting to another tool.

[SCREENSHOT: Claude analyzing a CSV — user asks "What were my top products by revenue?" and Claude returns a formatted table with product names, revenue, and percentages]

The before and after

To understand how big the difference is, imagine the same question answered two ways.

You ask: "What were my top 3 regions by revenue this quarter?"

Without Synap's data engine: Claude reads your entire sales.csv — all 10,000 rows, every column, every value. It loads the full file into its context, scrolls through thousands of lines of raw text, and tries to mentally sort and sum the numbers. This takes about 50,000 tokens, costs roughly the same as a long conversation, and sometimes the answer is wrong because scanning raw text is error-prone.

With Synap's data engine: Claude asks "what columns does this file have?" (answer: date, region, amount — 50 tokens), then runs a precise query: "sum amount by region, order by total, top 3" (answer: three rows — 80 tokens). Total: around 300 tokens. The answer is exact, every time.

Same question. Same file. One approach costs 150 times more than the other and is less reliable.

[SCREENSHOT: Side-by-side comparison — left panel shows Claude reading raw CSV text with thousands of lines visible, right panel shows Claude receiving a clean 3-row query result. Labels indicate "50,000 tokens" vs "300 tokens"]

Working with multiple files

The real power shows up when you need to combine information from more than one file. Things that would be nearly impossible by reading files as text become simple questions.

• "Join my customer list with my orders file and show me who hasn't ordered in 90 days."
• "Compare this month's sales report with last month's — what changed the most?"
• "Which products appear in my inventory file but have zero entries in my sales file?"
• "Combine my three regional reports into one view and rank by growth rate."

Each of these involves reading from multiple files, matching data between them, and computing results. Without the data engine, Claude would need to read both entire files and try to cross-reference them in its head — slow, expensive, and unreliable at scale. With the engine, it writes a single query that joins the files and gets back exactly the answer.

[SCREENSHOT: Claude joining two files — user asks "combine my customer list with orders and find customers who haven't ordered in 90 days" and Claude returns a table showing customer names and their last order dates]

What happens behind the scenes

You don't need to know this to use the feature — it works automatically. But if you're curious about why the answers are so fast and precise, here's what happens when you ask Claude a question about a data file.

Claude follows a three-step process, entirely on its own:

1. Discover. Claude asks: "What data files exist in this mind?" Synap returns a list of all CSV, TSV, and Parquet files it finds, along with their locations and sizes. This is like a librarian checking the catalog.

2. Inspect. Claude picks the file that looks relevant and asks: "What columns does this file have? How many rows? Show me a few examples." Synap returns the schema — column names, data types, a sample of the first few rows. This is like reading the table of contents.

3. Query. Now Claude knows exactly what it's working with. It writes a precise question — not "read the whole file" but "give me the sum of column X grouped by column Y, sorted by total, top 10." Synap runs that query locally and returns just the answer. This is like opening a book directly to the right page.

The entire process typically uses under 500 tokens. Reading the same file as text would use 50,000 or more. That's why Synap can analyze your data faster, cheaper, and more accurately than any approach that treats files as raw text.

[SCREENSHOT: A three-step visual — Step 1: Claude sees "3 data files found", Step 2: Claude sees column names and types for sales.csv, Step 3: Claude receives a clean query result with 5 rows]

The numbers

Before: reading a 10,000-row spreadsheet used around 50,000 tokens. After: querying it uses around 300 tokens. That's 99% less — not a rounding error.

In practice, this means faster answers and a lower cost per question. It also means you can ask more complex questions, like ones that compare two files or look for trends across many months, without running into limits.

What files work

The data engine understands CSV, TSV, Parquet, and JSONL files. If you can open a file in a spreadsheet, Synap can analyze it.

You don't need to prepare the files in any special way. Just put them in your mind and start asking questions.

[SCREENSHOT: The list of data files Claude discovered in a mind — showing file paths, formats (CSV, TSV), and sizes]

Your data stays private

Everything runs locally on your computer. The data engine is embedded inside the app — there's no external service, no cloud processing step, nothing leaving your machine.

When Claude queries your spreadsheet, it happens entirely on your device. Your data never moves.

[SCREENSHOT: Claude describing a data file — showing column names, data types, row count, and 3 sample rows]

No setup required

Update to Synap v1.0.5 or later and the data engine is already there. It activates automatically when you open your mind. Claude discovers your data files on its own and knows how to work with them.

You don't flip a switch, you don't enable a setting. The first time you ask Claude a question about a CSV in your mind, it just works.