perk11/pipod

Pipod is a script that runs AI coding agents like the pi coding agent, Claude Code, or OpenAI Codex CLI inside an isolated Docker container instead of directly on your computer. Each project directory gets its own persistent container, and the current directory is automatically shared into that container as a workspace folder. You can optionally turn off internet access for extra isolation, while still allowing the agent to reach services running on your own machine. The main point of running an agent this way is safety. If a coding agent is allowed to run shell commands or edit files, doing that inside a container limits the damage it can cause to the rest of your computer, compared to tools that only try to filter which commands are allowed or blocked. The README is honest that this is not a perfect sandbox, since the agent could still change files inside the mounted project folder or attempt to escape the container, but it describes this as a much stronger approach than command filtering alone. Getting started involves installing Docker, cloning the repository, and adding a shell alias pointing to the pipod script. From inside a project folder, running the pipod command launches the pi coding agent in its container. Passing claude or codex as an argument instead runs Claude Code or the Codex CLI in their own separate containers and Docker images, so installed packages for each agent do not mix. Extra flags let you drop into a plain shell inside the container to install project dependencies, force an isolated network with no outside access, or recreate the container from scratch. Configuration for each agent, such as Claude Code's settings and login credentials, is shared across all your projects by default, while session data and history for each individual project stay separate so different workspaces do not overwrite each other's work. The first time you log in to an agent inside a container, your existing host credentials are reused automatically where possible, or you can log in interactively once and it will persist for future sessions. The README includes detailed tables showing exactly which folders on your computer map to which paths inside each agent's container.