Code Interpreter Tool¶

The Code Interpreter tool executes arbitrary code in a sandboxed cloud VM powered by E2B. Code runs in full isolation from the host system, making this the safest option for untrusted or AI-generated code execution.

Quick Reference¶

Property	Value
Node name	`tools/code-interpreter`
Version	0.1.0
Library	`@e2b/code-interpreter`
Actions	execute, installPackages
Required config	`apiKey` (or `E2B_API_KEY` env var)
Tags	code, sandbox, e2b, interpreter, python, tools, agentic

Supported Languages¶

Language	Identifier	Package Manager
Python	`python`	pip
JavaScript	`javascript`	npm
TypeScript	`typescript`	npm
R	`r`	install.packages()
Java	`java`	Maven/Gradle (manual)
Go	`go`	go get
Ruby	`ruby`	gem
PHP	`php`	composer
Bash	`bash`	apt-get

Actions¶

execute¶

Execute code in the specified language.

Parameter	Type	Required	Description
`code`	string	Yes	Source code to execute
`language`	enum	No	Language identifier (default: `python`)
`timeout`	integer	No	Execution timeout in ms (default: 60000)
`files`	array	No	Files to upload before execution (see below)

installPackages¶

Install packages in the sandbox environment using the language's native package manager.

Parameter	Type	Required	Description
`packages`	string[]	Yes	Package names to install
`language`	enum	No	Determines which package manager to use (default: `python`)

Note

Java packages cannot be installed dynamically. The tool prints a notice directing users to Maven or Gradle.

File Upload¶

You can upload files into the sandbox before code execution:

{
  action: 'execute',
  code: 'import pandas as pd; df = pd.read_csv("data.csv"); print(df.describe())',
  language: 'python',
  files: [
    { name: 'data.csv', content: 'name,value\nalice,10\nbob,20' }
  ]
}

Each file object has:

Field	Type	Description
`name`	string	Filename (written to the sandbox working directory)
`content`	string	File content as a string

Output Schema¶

Field	Type	Description
`stdout`	string	Standard output (may be truncated)
`stderr`	string	Standard error (may be truncated)
`results`	array	Structured execution results (e.g. rich output from Jupyter-style cells)
`error`	string	Error message if execution failed
`language`	string	Language that was used
`success`	boolean	`true` if no error occurred

Configuration Reference¶

Property	Type	Default	Description
`apiKey`	string	(none)	E2B API key. Falls back to `E2B_API_KEY` environment variable.
`sandboxTimeout`	integer	`300000` (5 min)	Sandbox lifetime in milliseconds.
`maxOutputLines`	integer	`500`	Maximum number of output lines before truncation.

Safety¶

Full VM isolation¶

Code executes inside an E2B cloud sandbox -- a dedicated virtual machine that is completely isolated from the FlowForge host. There is no filesystem, network, or process sharing between the sandbox and the host.

Sandbox timeout¶

The sandbox is automatically destroyed after sandboxTimeout milliseconds, even if execution is still running. This prevents runaway processes from consuming resources indefinitely.

Output truncation¶

Both stdout and stderr are truncated to maxOutputLines lines. Truncated output includes a count of dropped lines.

Cleanup¶

The sandbox is always killed in a finally block, ensuring resources are released even when execution throws an error.

Warning

An E2B API key is required. Without it, the tool throws immediately. Get a key at e2b.dev.

Usage Example¶

import { codeInterpreterNode } from '@flowforgejs/nodes';

const workflow = {
  nodes: [
    {
      id: 'analyze-data',
      node: codeInterpreterNode,
      config: {
        apiKey: process.env.E2B_API_KEY,
        sandboxTimeout: 120_000,
        maxOutputLines: 200,
      },
      input: {
        action: 'execute',
        language: 'python',
        code: `
import pandas as pd
import json

df = pd.read_csv("sales.csv")
summary = {
    "total_revenue": float(df["revenue"].sum()),
    "avg_order_value": float(df["revenue"].mean()),
    "top_product": df.groupby("product")["revenue"].sum().idxmax(),
}
print(json.dumps(summary, indent=2))
        `.trim(),
        files: [
          { name: 'sales.csv', content: 'product,revenue\nwidget,100\ngadget,250\nwidget,150' },
        ],
      },
    },
  ],
};

Tip

For multi-step data analysis, install packages first with installPackages, then run your analysis with execute. The sandbox persists between calls within the same sandbox lifetime.