Agentspace runs parallel Claude/Codex agents in isolated Docker containers
Agentspace is a CLI tool by Notch123 that spawns Claude Code and Codex agents into Docker containers for long-running, parallel, isolated agent sessions — without tmux, git worktrees, or SSH disconnect risk.
Score breakdown
Agentspace enforces agent isolation and git-write restrictions at the container image level, removing the need to manually manage tmux sessions or git worktrees for parallel, long-running AI coding agent workflows.
- 01Built by Notch123 as a CLI to automate spawning Claude Code and Codex agents into Docker containers
- 02Agents run in YOLO mode by default inside isolated, volume-mounted containers
- 03Git commit and push are blocked at the image level via `PreToolUse` hooks (Claude Code) and exec policy rules (Codex), even under `--dangerously-skip-permissions`
Agentspace is a CLI tool that wraps Docker to provide long-running, parallel AI coding agent sessions without the shell-behavior quirks of terminal multiplexers or the awkwardness of git worktrees. Each agent is spawned into its own Docker container with an isolated volume-mounted copy of the codebase, allowing many tasks to run concurrently without risk to the host machine. Agents run in YOLO mode by default, and session persistence is handled by Docker itself — containers keep running through SSH disconnects and can be reattached at any time with `agentspace my-task attach`, using the standard Docker detach sequence (`Ctrl-P Ctrl-Q`).
A dedicated preview command launches a separate container attached to the same volume for real-time inspection of agent changes.
A dedicated preview command launches a separate container attached to the same volume for real-time inspection of agent changes. Docker-in-Docker support is available via a sidecar using `--privileged` mode, though the author notes this is not a sandbox against actively hostile code. Git commit and push operations are blocked at the image level — through `PreToolUse` hooks for Claude Code and exec policy rules for Codex — keeping the user in control of what reaches the repo. Work exits the container via a git remote, after which it appears as a normal branch for standard PR review. The base image is `node:24-slim`, so Node/TypeScript works out of the box; other runtimes require a custom Dockerfile pointed to via the `AGENTSPACE_IMAGE` environment variable. Currently supports Claude Code and Codex, with Opencode and Pi listed as future targets. Windows support is unverified, and the author notes the project was built in roughly a week with some rough edges remaining.
Key facts
- 01Built by Notch123 as a CLI to automate spawning Claude Code and Codex agents into Docker containers
- 02Agents run in YOLO mode by default inside isolated, volume-mounted containers
- 03Git commit and push are blocked at the image level via `PreToolUse` hooks (Claude Code) and exec policy rules (Codex), even under `--dangerously-skip-permissions`
- 04Sessions survive SSH disconnects; detach with Ctrl-P Ctrl-Q and reattach with `agentspace my-task attach`
- 05Docker-in-Docker sidecar support uses `--privileged` mode and is not a sandbox against actively hostile code
- 06Base image is `node:24-slim`; other runtimes require a custom Dockerfile via `AGENTSPACE_IMAGE`
- 07Supports Claude Code and Codex today; Opencode and Pi are planned; Windows is unverified
Topics
Summary and scoring are generated automatically from the original article. We always link back to the publisher and never republish images or paywalled content. Last processed Jun 18, 2026 · 10:40 UTC. How this works →