<?xml version="1.0" encoding="utf-8" standalone="yes"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title>Langgraph on Anshul Patel</title><link>http://www.anshulpatel.in/tags/langgraph/</link><description>Recent content in Langgraph on Anshul Patel</description><generator>Hugo</generator><language>en</language><lastBuildDate>Tue, 26 May 2026 00:00:00 +0000</lastBuildDate><atom:link href="http://www.anshulpatel.in/tags/langgraph/index.xml" rel="self" type="application/rss+xml"/><item><title>Simple AI Agent Sandbox: Ephemeral Docker Isolation for Every AI Session</title><link>http://www.anshulpatel.in/posts/simple_ai_agent_sandbox/</link><pubDate>Tue, 26 May 2026 00:00:00 +0000</pubDate><guid>http://www.anshulpatel.in/posts/simple_ai_agent_sandbox/</guid><description>&lt;p&gt;AI agents are exciting. Give a capable LLM the ability to run code, write files,
and execute shell commands, and suddenly you have a system that can actually
&lt;em&gt;do&lt;/em&gt; things rather than just talk about them. But that power comes with a sharp
edge: &lt;strong&gt;an agent that can run arbitrary shell commands can do a lot of damage if
it goes sideways.&lt;/strong&gt;&lt;/p&gt;
&lt;p&gt;The typical response is to add guardrails at the prompt level: &lt;em&gt;&amp;ldquo;don&amp;rsquo;t delete
files&amp;rdquo;&lt;/em&gt;, &lt;em&gt;&amp;ldquo;only modify files in this directory&amp;rdquo;&lt;/em&gt;. This works until it doesn&amp;rsquo;t.
Prompts are soft constraints. A Docker container is a hard one.&lt;/p&gt;
&lt;p&gt;That&amp;rsquo;s the idea behind
&lt;a href="https://github.com/anshulpatel25/simple-ai-agent-sandbox" class="external-link" target="_blank" rel="noopener"&gt;&lt;strong&gt;simple-ai-agent-sandbox&lt;/strong&gt;&lt;/a&gt;:
&lt;em&gt;every agent session gets its own fresh, ephemeral Ubuntu container&lt;/em&gt;. When the
session ends, the container is destroyed. Clean slate, every time.&lt;/p&gt;</description></item></channel></rss>