Contained Intelligence: Secure Sandboxing for Gen AI Code
summary
Demonstrates a working pipeline: Claude generates Python that is uploaded and executed in a Daytona sandbox via Daytona MCP, producing the correct SHA output instead of a hallucinated guess. Shows CLI/SDK examples (TypeScript, Python), web terminal/SSH access, Docker image support and port previewing, and create sandbox in <3s during the demo. Describes scale and uses with concrete numbers: thousands of parallel sandboxes for RL training, file uploads, git ops, and JSON exit-code results used as agent context.
AI-written summary · about
chapters (15)
- 0:00Introduction and speaker intro
- 1:04What is an AI agent?
- 3:13Limitations of LLMs with code
- 4:38Introducing tools for AI agents
- 7:03Why sandboxing is essential
- 8:47LLM vs sandboxing workflow
- 10:15Connecting AI agents to tools
- 12:44MCP protocol overview
- 13:48Configuring Daytona MCP
- 15:13Demo: Executing code in sandbox
- 17:29Coding sandboxing yourself
- 19:58Sandbox features and capabilities
- 21:24Use cases beyond code execution
- 22:51Deployment options and open source
- 23:54Closing and Q&A
related talks
01h 48mHow to add secure code interpreting in your AI app
25mPlease Sandbox your AI Agents
38mWhy, and how you need to sandbox AI-Generated Code?
12mDocker Sandboxes for AI
34mZero Secrets: Building Secure Infrastructure for AI Agents
01h 03mBuilding Secure Infrastructure for Productive AI Agents
32mLLMs Behind Bars: Sandboxes at Scale for AI on a Short Leash
15mSandboxes in Production