AI coding agents work best when you let them run commands, install packages and edit files without confirming every step. That freedom is also what makes them dangerous on a developer machine: a wrong delete, a leaked token, or a prompt injection that tricks your agent into something you never asked it to do.

Eclipse Enclave is an open source runtime that puts each agent session into its own container, with a separate filesystem and network stack. The agent sees only the paths you mount, usually the project worktree, and reaches only the domains you allow.

This webinar introduces Enclave and its installation, then uses it for real coding work with agents like Claude Code and Codex CLI. The second part covers extending it: how to add a harness that Enclave does not ship out of the box, for example OpenClaw or a DeepSeek-based CLI, and run it under the same isolation.

OCX Webinar Week - Isolating AI Agents with Eclipse Enclave: Claude Code, Codex, OpenClaw and more