What it does

VibePod is a command-line tool, invoked as vp, that starts AI coding agents inside their own Docker or Podman containers. The basic use is vp run followed by an agent name, with no required configuration; optional YAML covers custom setups. The README lists many supported agents, including Claude, Gemini, Codex, OpenCode, Copilot, Auggie, Pi and Qwen. You can register any OpenAI- or Anthropic-compatible endpoint or local model server once and route agents to it, install reusable prompt recipes as skills, and add per-project Dockerfile overlays that build into cached image layers. A flag turns the container into an Agent Client Protocol adapter, so an editor such as Zed can use it in its AI panel. While agents run, VibePod collects usage and HTTP traffic metrics and serves them in a built-in dashboard that includes Claude token metrics and a side-by-side agent comparison. The README says the data stays on your machine. It installs through pip, Homebrew or conda-forge and is MIT licensed.

Capabilities

Start an agent in its own container with a single vp run command
Supports many agents, including Claude, Codex, Gemini, OpenCode and Pi
Register OpenAI- or Anthropic-compatible endpoints and local model servers
Local dashboard of HTTP traffic, usage and token metrics
Side-by-side agent comparison
Installs through pip, Homebrew or conda-forge

When to use it

You want coding agents to run in isolation from the rest of your machine
You want to compare several agents on usage and traffic
You want your editor's AI panel to use a containerized agent