Zero-config local agents
Built-in local metrics, HTTP traffic tracking, and an analytics dashboard to see what your agents do. Zero config — just vp run claude in Docker or Podman.
Install VibePod once. Run Claude, Gemini, Codex, Devstral, Copilot, Auggie, OpenCode, Pi, Antigravity, Tau, Jcode, Freebuff, Qwen Code, or the DeepSeek Harness in an isolated container with a single command. Add reusable prompt recipes with vp skills, and add global or project YAML only when you want shared defaults, custom images, env vars, init scripts, or LLM routing.
Latest News
Follow VibePod releases and cross-repository work across the CLI, agent images, dashboard, proxy, website, and packaging repositories.
The Web-UI-first DeepSeek Harness joins the agent matrix as dsh, vp proxy filter allows or blocks hosts per credential profile, and vp run -p publishes a port for a single session.
Keep multiple credential sets per agent and switch with --profile, run Alibaba's Qwen Code as qwen and Codebuff's Freebuff as freebuff, and keep working offline with the local image fallback.
Projects extend agent images with committed Dockerfile fragments, agents report live state to herdr panes, containers publish ports via per-agent config, and Jcode joins as jcode.
Zero config by default
VibePod ships with major coding agents pre-configured. Start immediately, then add global or project YAML later for overrides your team wants to share.
| Manual setup | VibePod |
|---|---|
| Write YAML before first run | Skip it; defaults are built in |
| Choose and configure a model upfront | Use agent defaults, or set LLM routing later |
| Wire up tools and servers | Optional env vars, init scripts, and custom images |
| Share project defaults | vp config init |
| Run your agent | vp run claude |
Quickstart
Install from PyPI, run any supported agent with no project config, then start the dashboard when you want local metrics.
python -m pip install vibepod
vp run claude
vp logs startTip: pin versions in automation with python -m pip install vibepod==<version>.
$ vp version
VibePod CLI: 0.21.0
Python: 3.12.3
Docker: 28.1.1
$ vp run claude
Starting claude with image vibepod/claude:latestHow it works
Run vp list to see which agents are available in your local setup.
Note: vp list also shows active workspaces.
Start work with vp run claude, then switch agents by rerunning the command with a different name.
Note: the current folder is used as the active agent workspace.
Run vp logs start to launch the local dashboard. It shows HTTP traffic, usage metrics, and lets you compare agents side by side.
Core features
No project file is needed to start. Claude, Gemini, Codex, Devstral, Copilot, Auggie, OpenCode, Pi, Antigravity, Tau, Jcode, Freebuff, Qwen Code, and the DeepSeek Harness are ready out of the box.
Proof point: run an agent with vp run <agent>.
Add .vibepod/config.yaml only when you want team defaults for images, env vars, init scripts, proxy, logging, or LLM settings.
Proof point: create a starter with vp config init.
Track HTTP traffic, usage, and activity per agent from a browser-based dashboard.
Proof point: launch it with vp logs start.
VibePod can start a local mitmproxy container alongside agents and write captured HTTP(S) traffic to SQLite.
Proof point: disable it with VP_PROXY_ENABLED=false.
Install prompt recipes per user or per project, then mount them into supported agents automatically.
Proof point: vp skills add github:vibepod/vibepod-skills//skills/researcher.
Keep long-running containers alive after a terminal closes, then reconnect to a managed container when needed.
Proof point: vp attach <container>.
Override an agent image for a one-off run or persist it in config, and run idempotent shell commands before startup.
Proof point: set VP_IMAGE_CLAUDE=myorg/claude:dev.
Route Claude or Codex to Ollama, vLLM, or another compatible endpoint with config or runtime environment variables.
Proof point: set VP_LLM_ENABLED=true.
When a workspace has Compose files, VibePod can connect an agent container to the right Docker or Podman network.
Proof point: use vp run claude --network my-net.
Why VibePod?
| Workflow need | Manual setup | VibePod |
|---|---|---|
| Agent configuration | Write YAML before first run | Pre-configured; YAML is optional |
| Switch agents | Rewrite config and rerun | vp run <agent> |
| Container isolation | Manual Docker or Podman setup | Built in |
| Project defaults | Hand-maintain setup docs | Commit .vibepod/config.yaml |
| Usage tracking | Bring your own tooling | Local dashboard |
| HTTP traffic logging | Manual proxy setup | Built in |
| Reusable prompts | Copy instructions between tools manually | vp skills add |
| Agent comparison | Manual notes | Side-by-side dashboard views |
| Data privacy | Depends on vendor and tooling | Metrics stored locally |
Ecosystem
Everything is MIT-licensed and open on GitHub — from the CLI to the agent images built and published automatically via CI.
The main CLI (vp). Install via pip, run any agent with a single command in an isolated container.
Python
Container definitions for all supported agents, built and published to Docker Hub automatically via CI.
JavaScript
HTTP proxy that captures all outbound traffic from agent containers and logs it to SQLite.
Python
Analytics dashboard for browsing captured metrics and logs via Datasette.
Python
Curated reusable skills that can be installed with vp skills for supported agents.
Markdown
Official Homebrew tap, so macOS and Linux users can install the CLI with a single brew command.
Ruby
Product screenshots
Included screenshots cover the analytics dashboard plus Auggie, Antigravity, Claude Code, Freebuff, GitHub Copilot, DeepSeek Harness, Gemini, Jcode, Mistral, OpenAI Codex, OpenCode, Pi, Qwen Code, and Tau.
FAQ
Install from PyPI using python -m pip install vibepod, then use the README for command examples.
No. VibePod comes pre-configured for Claude, Gemini, Codex, Devstral, Copilot, Auggie, OpenCode, Pi, Antigravity, Tau, Jcode, Freebuff, Qwen Code, and the DeepSeek Harness. Run vp run <agent> and it starts from the default image.
Yes. YAML is optional. Use global config at ~/.config/vibepod/config.yaml or project config at .vibepod/config.yaml to override defaults, share team settings, set custom images, inject env vars, run init scripts, or configure LLM routing.
The dashboard tracks HTTP traffic, usage over time, and activity per agent. Start it with vp logs start; it opens in your browser and stores data locally.
Yes. The core workflow is local CLI usage so you can run and switch coding agents from your machine in isolated Docker or Podman containers.
Either. VibePod auto-detects rootless Podman and applies the right user-namespace settings, so vp run <agent> works the same way under Docker or Podman. See the 0.16 release notes and the Podman quickstart.
Yes. Run vp --install-completion to enable auto-complete and speed up command entry and agent selection.
The current working directory where commands are run is used as the active agent workspace.
Yes. The same CLI flow used locally can be scripted for CI/CD jobs.
VibePod currently supports Claude, Gemini, OpenCode, Devstral, Auggie, GitHub Copilot, OpenAI Codex, Pi, Antigravity, Tau, Jcode, Freebuff, Qwen Code, and the DeepSeek Harness.
Use the GitHub issues page to follow changes and request features.
Install from PyPI, start a preconfigured agent, and add YAML only when your team needs shared defaults or deeper customization.
Enjoying VibePod? Star it on GitHub — it helps other developers find the project. 74 and counting.