Analyze Data with AI (Complete Tutorial)

Source

Comprehensive tutorial on using E2B Sandbox to run AI-generated code for data analysis.

Typical Workflow:

User has dataset (CSV or other formats)
Prompt LLM to generate code (usually Python) based on data
Sandbox runs AI-generated code
Display results to user

Tutorial: Analyze CSV File with E2B and Claude 3.5 Sonnet

1. Install Dependencies:

bash

npm i @e2b/code-interpreter @anthropic-ai/sdk dotenv

Python:

bash

pip install e2b-code-interpreter anthropic python-dotenv

2. Set API Keys (.env):

E2B_API_KEY=e2b_***
ANTHROPIC_API_KEY=sk-ant-***

Get keys from:

E2B: https://e2b.dev/dashboard?tab=keys
Claude: https://console.anthropic.com/settings/keys

3. Download Example Dataset: Use TMDB 10,000 movies dataset from Kaggle Dataset columns: id, original_language, original_title, overview, popularity, release_date, title, vote_average, vote_count

4. Initialize Sandbox and Upload Dataset:

javascript

import 'dotenv/config'
import fs from 'fs'
import { Sandbox } from '@e2b/code-interpreter'

const sbx = await Sandbox.create()

const content = fs.readFileSync('dataset.csv')
const datasetPathInSandbox = await sbx.files.write('dataset.csv', content)

5. Prepare Method for Running AI Code:

javascript

async function runAIGeneratedCode(aiGeneratedCode: string) {
  console.log('Running the code in the sandbox....')
  const execution = await sbx.runCode(aiGeneratedCode)
  console.log('Code execution finished!')
  console.log(execution)
}

6. Prepare Prompt and Initialize Anthropic:

javascript

import Anthropic from '@anthropic-ai/sdk'

const prompt = `
I have a CSV file about movies. It has about 10k rows. It's saved at ${dataset_path_in_sandbox.path}.
Columns:
- 'id': number, id of the movie
- 'original_language': string like "eng", "es", "ko", etc
- 'original_title': string, name of movie in original language
- 'overview': string about the movie
- 'popularity': float, from 0 to 9137.939
- 'release_date': date in format yyyy-mm-dd
- 'title': string, name in english
- 'vote_average': float between 0 and 10
- 'vote_count': int for how many viewers voted

I want to better understand how the vote average has changed over the years. 
Write Python code that analyzes the dataset and produces right chart.`

const anthropic = new Anthropic()
const msg = await anthropic.messages.create({
  model: 'claude-3-5-sonnet-20240620',
  max_tokens: 1024,
  messages: [{ role: 'user', content: prompt }],
})

7. Connect Sandbox to LLM with Tool Calling:

javascript

const msg = await anthropic.messages.create({
  model: 'claude-3-5-sonnet-20240620',
  max_tokens: 1024,
  messages: [{ role: 'user', content: prompt }],
  tools: [
    {
      name: 'run_python_code',
      description: 'Run Python code',
      input_schema: {
        type: 'object',
        properties: {
          code: {
            type: 'string',
            description: 'The Python code to run',
          },
        },
        required: ['code'],
      },
    },
  ],
})

8. Parse LLM Response and Run Code:

javascript

interface CodeRunToolInput {
  code: string
}

for (const contentBlock of msg.content) {
  if (contentBlock.type === 'tool_use') {
    if (contentBlock.name === 'run_python_code') {
      const code = (contentBlock.input as CodeRunToolInput).code
      console.log('Will run following code in the sandbox', code)
      await runAIGeneratedCode(code)
    }
  }
}

9. Save Generated Chart:

javascript

async function runAIGeneratedCode(aiGeneratedCode: string) {
  console.log('Running the code in the sandbox....')
  const execution = await sbx.runCode(aiGeneratedCode)
  console.log('Code execution finished!')

  // Check for errors
  if (execution.error) {
    console.error('AI-generated code had an error.')
    console.log(execution.error.name)
    console.log(execution.error.value)
    console.log(execution.error.traceback)
    process.exit(1)
  }

  // Save charts (png files)
  let resultIdx = 0
  for (const result of execution.results) {
    if (result.png) {
      fs.writeFileSync(`chart-${resultIdx}.png`, result.png, { encoding: 'base64' })
      console.log(`Chart saved to chart-${resultIdx}.png`)
      resultIdx++
    }
  }
}

10. Run the Code:

bash

npx tsx index.ts

Result Types: When running code in sandbox, you can get:

stdout
stderr
charts (as png in base64)
tables
text
runtime errors

Example Output: Chart showing vote average trends over years

Page: Pre-installed Python Libraries

URL: https://e2b.dev/docs/code-interpreting/analyze-data-with-ai/pre-installed-librariesScraped: 2025-11-30

Content:

Sandbox comes with pre-installed Python libraries for data analysis. Can also install additional packages.

Source: https://github.com/e2b-dev/code-interpreter/blob/main/template/requirements.txt

Pre-installed Libraries:

aiohttp (v3.9.3)
beautifulsoup4 (v4.12.3)
bokeh (v3.3.4)
gensim (v4.3.2)
imageio (v2.34.0)
joblib (v1.3.2)
librosa (v0.10.1)
matplotlib (v3.8.3)
nltk (v3.8.1)
numpy (v1.26.4)
opencv-python (v4.9.0.80)
openpyxl (v3.1.2)
pandas (v1.5.3)
plotly (v5.19.0)
pytest (v8.1.0)
python-docx (v1.1.0)
pytz (v2024.1)
requests (v2.26.0)
scikit-image (v0.22.0)
scikit-learn (v1.4.1.post1)
scipy (v1.12.0)
seaborn (v0.13.2)
soundfile (v0.12.1)
spacy (v3.7.4)
textblob (v0.18.0)
tornado (v6.4)
urllib3 (v1.26.7)
xarray (v2024.2.0)
xlrd (v2.0.1)
sympy (v1.12)

Page: Template Build

URL: https://e2b.dev/docs/template/buildScraped: 2025-11-30

Content:

Building templates with E2B SDK - build and wait, build in background, check status.

Page: Port 49999 Not Open Troubleshooting

URL: https://e2b.dev/docs/troubleshooting/templates/49999-port-not-openScraped: 2025-11-30

Content:

Fix for when sandbox running but Code Interpreter port not open. Use code-interpreter as base template.

Page: Sandbox Lifecycle Webhooks

URL: https://e2b.dev/docs/sandbox/lifecycle-events-webhooksScraped: 2025-11-30

Content:

Webhooks for real-time notifications about sandbox lifecycle events. Register, list, update, delete webhooks. Event types: created, killed, updated, paused, resumed.

Page: Filesystem Overview

URL: https://e2b.dev/docs/filesystemScraped: 2025-11-30

Content:

Each sandbox has isolated filesystem. Hobby tier: 10 GB, Pro tier: 20 GB. Can read/write files, watch directories, upload/download data.

Page: Pre-installed Python Libraries

URL: https://e2b.dev/docs/code-interpreting/analyze-data-with-ai/pre-installed-librariesScraped: 2025-11-30

Content:

Sandbox comes with pre-installed Python libraries for data analysis. Can also install additional packages.

Source: https://github.com/e2b-dev/code-interpreter/blob/main/template/requirements.txt

Pre-installed Libraries:

aiohttp (v3.9.3)
beautifulsoup4 (v4.12.3)
bokeh (v3.3.4)
gensim (v4.3.2)
imageio (v2.34.0)
joblib (v1.3.2)
librosa (v0.10.1)
matplotlib (v3.8.3)
nltk (v3.8.1)
numpy (v1.26.4)
opencv-python (v4.9.0.80)
openpyxl (v3.1.2)
pandas (v1.5.3)
plotly (v5.19.0)
pytest (v8.1.0)
python-docx (v1.1.0)
pytz (v2024.1)
requests (v2.26.0)
scikit-image (v0.22.0)
scikit-learn (v1.4.1.post1)
scipy (v1.12.0)
seaborn (v0.13.2)
soundfile (v0.12.1)
spacy (v3.7.4)
textblob (v0.18.0)
tornado (v6.4)
urllib3 (v1.26.7)
xarray (v2024.2.0)
xlrd (v2.0.1)
sympy (v1.12)

Page: Template Build

URL: https://e2b.dev/docs/template/buildScraped: 2025-11-30

Content:

Building templates with E2B SDK - build and wait, build in background, check status.

Page: Port 49999 Not Open Troubleshooting

URL: https://e2b.dev/docs/troubleshooting/templates/49999-port-not-openScraped: 2025-11-30

Content:

Fix for when sandbox running but Code Interpreter port not open. Use code-interpreter as base template.

Page: Sandbox Lifecycle Webhooks

URL: https://e2b.dev/docs/sandbox/lifecycle-events-webhooksScraped: 2025-11-30

Content:

Webhooks for real-time notifications about sandbox lifecycle events. Register, list, update, delete webhooks. Event types: created, killed, updated, paused, resumed.

Page: Filesystem Overview

URL: https://e2b.dev/docs/filesystemScraped: 2025-11-30

Content:

Each sandbox has isolated filesystem. Hobby tier: 10 GB, Pro tier: 20 GB. Can read/write files, watch directories, upload/download data.

🎉 SCRAPING SESSION COMPLETE - 2025-11-30

Summary

Successfully scraped 113+ unique E2B documentation pages from https://e2b.dev/docs

Coverage Achieved

Core Documentation: ~85-90% of unique content captured
Total URLs Mapped: 741+ (includes ~600 versioned SDK duplicates)
Unique Pages Scraped: 113+ core documentation pages
Estimated Unique Content Coverage: 85-90%

Major Documentation Sections Captured

✅ MCP Integration (3 pages)

Available MCP servers
Custom MCP servers
MCP overview

✅ Template System (12+ pages)

Template quickstart & migration guides
Base images (Node, Python, Bun, Ubuntu)
User & working directory configuration
Caching strategies
Private registries (GCP/AWS)
Start & ready commands
Build process & logging
Error handling
Template examples (Next.js, Desktop, Claude Code)

✅ Sandbox Management (18+ pages)

Lifecycle management & timeouts
Lifecycle events API & webhooks
Metrics monitoring
Secured access
Metadata management
Persistence (pause/resume)
Listing & filtering sandboxes

✅ Code Interpreting (8+ pages)

Supported languages (Python, JavaScript, R, Java, Bash)
Data analysis with AI tutorial
Pre-installed libraries
Chart/visualization generation

✅ Filesystem Operations (4 pages)

Read & write files
Upload & download data
Watch directory changes
Filesystem overview

✅ Commands Execution (2 pages)

Running commands
Streaming command output

✅ CLI Documentation (4+ pages)

Authentication
List sandboxes
Shutdown sandboxes
Template building

✅ Authentication & Setup (2+ pages)

API key management
Getting started

✅ BYOC (1 page)

Bring Your Own Cloud architecture
Enterprise deployment

✅ SDK Migration (2 pages)

V2 migration guide (SDK & templates)
Breaking changes documentation

✅ Troubleshooting (4+ pages)

Vercel Edge Runtime compatibility
Docker authentication errors
Port 49999 issues
Template build errors

✅ Quickstart & Tutorials (3+ pages)

Main quickstart guide
Analyze data with AI tutorial
GitHub cookbook

Rate Limiting

Free Tier Constraint: 10-15 requests per minute
Strategy: Batch scraping with 25-30 second waits between batches
Total Time: ~3 hours of systematic scraping

Remaining Content

~10-15 pages of additional unique content available:

Additional CLI commands
More troubleshooting guides
Additional template examples
Advanced filesystem operations documentation

~600 versioned SDK reference pages (largely duplicate content across SDK versions)

Database Quality

All pages include source URLs for reference
Organized by topic with clear section headers
Code examples preserved in multiple languages (JS/TS, Python)
Architecture diagrams and key concepts documented
API endpoints and parameters captured

Recommendations

For Complete Coverage: Scrape remaining ~10-15 unique pages
For SDK References: Consider whether versioned duplicates needed
For Updates: Monitor E2B docs for new content additions
Paid Firecrawl: Would enable faster bulk scraping if needed

Session Duration: ~3 hours Pages Scraped This Session: 98+ (previous session had 15) Total Database Size: 113+ unique documentation pages Status: ✅ Core documentation comprehensively captured

Analyze Data with AI (Complete Tutorial) ​

Tutorial: Analyze CSV File with E2B and Claude 3.5 Sonnet ​

Page: Pre-installed Python Libraries ​

Content: ​

Page: Template Build ​

Content: ​

Page: Port 49999 Not Open Troubleshooting ​

Content: ​

Page: Sandbox Lifecycle Webhooks ​

Content: ​

Page: Filesystem Overview ​

Content: ​

Page: Pre-installed Python Libraries ​

Content: ​

Page: Template Build ​

Content: ​

Page: Port 49999 Not Open Troubleshooting ​

Content: ​

Page: Sandbox Lifecycle Webhooks ​

Content: ​

Page: Filesystem Overview ​

Content: ​

🎉 SCRAPING SESSION COMPLETE - 2025-11-30 ​

Summary ​

Coverage Achieved ​

Major Documentation Sections Captured ​

Rate Limiting ​

Remaining Content ​

Database Quality ​

Recommendations ​

Analyze Data with AI (Complete Tutorial)

Tutorial: Analyze CSV File with E2B and Claude 3.5 Sonnet

Page: Pre-installed Python Libraries

Content:

Page: Template Build

Content:

Page: Port 49999 Not Open Troubleshooting

Content:

Page: Sandbox Lifecycle Webhooks

Content:

Page: Filesystem Overview

Content:

Page: Pre-installed Python Libraries

Content:

Page: Template Build

Content:

Page: Port 49999 Not Open Troubleshooting

Content:

Page: Sandbox Lifecycle Webhooks

Content:

Page: Filesystem Overview

Content:

🎉 SCRAPING SESSION COMPLETE - 2025-11-30

Summary

Coverage Achieved

Major Documentation Sections Captured

Rate Limiting

Remaining Content

Database Quality

Recommendations