AI Data Analysis

Build AI data analysis tools where users upload their data and ask questions. Each user's data stays in their own isolated environment — private, persistent, searchable.

The problem

Users upload CSVs, connect databases, ask questions about their data. Your agent needs to run code on that data — pandas, SQL, charting. But you can't run untrusted code in a shared environment. Each user needs isolation, and their data needs to persist across sessions.

How OnCell solves it

Each user gets their own sandboxed environment. They upload files to their own storage. Your agent runs code inside — pandas, matplotlib, SQL queries — with direct access to the data at 0ms latency. No data leaves the sandbox.

const cell = await oncell.cells.create({
  customerId: "analyst-456",
  agent: dataAgentCode,
});

// User uploads data
await oncell.cells.writeFile(cell.id, "data/sales.csv", csvContent);

// Agent analyzes it
// Inside the agent:
ctx.shell("python -c 'import pandas as pd; df = pd.read_csv("/cell/work/data/sales.csv"); print(df.describe())'");
ctx.stream({ type: "insight", text: "Revenue increased 23% QoQ" });

What your agent gets

Per-user storage— each user's data is isolated. Upload CSVs, JSON, Excel files. Persistent across sessions.

Code execution — run Python, shell scripts, SQL. Full access to pandas, numpy, matplotlib.

Database — store analysis results, user preferences, cached computations.

Streaming — stream insights and chart data to the frontend as the agent discovers patterns.

Data privacy— each user's data is physically isolated on NVMe. No cross-user access possible.

Example architectures

Chat-with-your-data — user uploads a CSV, asks "what are the trends?", agent runs pandas and generates insights with charts.

AI spreadsheet — user connects a data source, agent builds formulas, generates reports, caches results.

Automated reporting— agent runs nightly analysis on each customer's data, generates PDF reports, stores in their cell.