Code Execute¶

The Code Execute sub-component provides a sandboxed code execution tool to agents. When connected to an agent, the agent's LLM can write and run Python or Bash code in an isolated subprocess with security restrictions and timeout enforcement.

Ports¶

Inputs¶

This component has no wired inputs. It operates as a LangChain tool that the agent invokes with arguments during its reasoning loop.

Outputs¶

Configuration¶

The Code Execute component accepts the following configuration in extra_config:

Usage¶

1. Add a Code Execute node from the Node Palette (Sub-components category)
2. Connect it to an agent node via the green diamond tools handle
3. Optionally configure the default language, timeout, and sandbox settings in Extra Config

The agent's LLM will then have a code_execute tool available. The tool accepts two parameters:

Execution environment¶

Code runs in a subprocess (not in the Pipelit process):

• Python -- written to a temp file in /tmp and executed with python3
• Bash -- written to a temp file in /tmp with #!/bin/bash and set -e, then executed with /bin/bash

Both languages run with a working directory of /tmp.

Sandbox mode¶

When sandbox is enabled (the default), two protections are applied:

1. Security pattern checking -- code is scanned for forbidden patterns before execution
2. Environment restriction -- the subprocess runs with a restricted PATH (/usr/bin:/bin:/usr/local/bin) and without HOME or USER environment variables

Forbidden patterns¶

The security scanner blocks code containing these patterns:

Python:

• import os, from os import, import subprocess, from subprocess import
• import shutil, from shutil import
• __import__(), eval(), exec(), compile()
• open() targeting /etc, /proc, /sys, /dev

Bash:

• rm -rf /, rm -rf ~, rm -rf $HOME
• dd if=... of=/dev/..., mkfs.
• Writes to /etc/ or /dev/
• Piped remote execution (curl ... | sh, wget ... | sh)
• chmod 777, chmod -R 777

Return format¶

The tool returns a multi-line string with:

stdout:
<captured stdout>
stderr:
<captured stderr>
exit_code: 0
result: <JSON-parsed last line, if applicable>

If the code times out, the output is: Execution timed out after N seconds.

If a security violation is detected, the output is: Security violation: <details>.

Example¶

An agent with Code Execute can perform computations, data processing, and system queries:

User: Calculate the first 20 Fibonacci numbers.
Agent: [calls code_execute(code="a,b=0,1\nfor _ in range(20):\n    print(a)\n    a,b=b,a+b")]
       → stdout:
         0
         1
         1
         2
         3
         5
         8
         13
         ...
         exit_code: 0
Agent: The first 20 Fibonacci numbers are: 0, 1, 1, 2, 3, 5, 8, 13, ...

flowchart LR
    T[Chat Trigger] --> A[Agent]
    M[AI Model] -.->|model| A
    CE[Code Execute] -.->|tool| A

    style CE fill:#10b981,color:white