Agentainer is one container image with four AI coding agents and a shared set of developer tools. You run it in a terminal on a laptop, a server or a Kubernetes cluster.
- Agents: Claude Code, Codex CLI, Antigravity CLI (
agy) and OpenCode v2. - Base: Ubuntu 26.04 LTS, pinned by digest, with apt packages from the Ubuntu snapshot service at a pinned date.
- Tools: git, gh, Go, Rust, Python with uv and pipx, Node.js, GCC/G++, CMake, Ninja, pkg-config, psql, kubectl, Helm, AWS CLI v2, abtop, ripgrep, jq, yq, shellcheck, trivy, gosec, gitleaks, poppler-utils (pdftotext, pdftoppm), ImageMagick, the Chromium libraries Playwright needs, the OpenSSH SFTP server and more.
/etc/agentainer/versions.jsonin the image lists every version. - Platforms:
linux/amd64andlinux/arm64. On amd64 the agents also start on CPUs without AVX2; CI checks this on every build.
Agents can modify mounted files, use supplied credentials and reach network services. See the security limits.
These docs describe the current source. Features merged after the newest release tag are only in a local build until the next release (release policy).
You need Podman or Docker. The launcher bin/agentainer is a POSIX shell script.
Set AGENTAINER_IMAGE to a published version tag; no latest tag exists before the first stable release (release versions). Or build locally with scripts/build.sh (RUNTIME=docker sh scripts/build.sh on a Docker host) and use localhost/agentainer:dev.
mkdir -p ~/.config/agentainer
cp env.example ~/.config/agentainer/env
chmod 600 ~/.config/agentainer/env # then fill in the keys you use
export AGENTAINER_IMAGE=ghcr.io/chadmayfield/agentainer:0.1.0-dev.20261024.1
bin/agentainer up ~/code # ~/code becomes /workspace
bin/agentainer attach # tmux session "main"Start an agent in a tmux window (claude, codex, agy or opencode). Close the terminal whenever you like. bin/agentainer attach brings you back to the same session with the agents still running.
The documentation index lists the same pages.
| Topic | Read |
|---|---|
Launcher commands, docker run and podman run, sudo, file ownership, upgrading |
docs/running.md |
| Signing in to each agent, which Claude login wins, GitHub tokens | docs/logins.md |
| Where each agent keeps settings, instructions and skills, status lines, MCP logins | docs/agents.md |
| Seeded shell, Git, tmux, Vim, EditorConfig, agent settings and the GitHub Dark colors | docs/defaults.md |
| Your own configs, bind-mounted homes, separate test homes and custom images | docs/customization.md |
The shared AGENTS.md, MCP servers, tools that agents install, OpenRouter |
docs/configuration.md |
| Python tests and linting, JavaScript tooling, native builds and PostgreSQL scripts | docs/project-tools.md |
| Optional Playwright setup, persistent browser downloads and execution checks | docs/browser-testing.md |
| Optional Go, Rust, Python and TypeScript language servers | docs/language-tooling.md |
| Optional agent plugins, token/context tools and documentation lookup | docs/plugins.md |
| Agent session monitoring with abtop | docs/monitoring.md |
| Kubernetes with ContainerSSH | docs/kubernetes.md |
| The Kelos variant image | docs/kelos.md |
| What the container isolates, agent permissions, vulnerability reports | docs/security.md |
| Pinned versions, scheduled updates, releases and signatures | docs/versions.md |
Offline diagnostics with agentainer doctor, error messages and fixes |
docs/troubleshooting.md |
To report a vulnerability, follow SECURITY.md. To contribute, read CONTRIBUTING.md.
MIT, see LICENSE. The image also contains software under other licenses.