Agentic OS base container image
This topic describes what the base image contains and the scenarios for which it is and is not suitable, explains its startup behavior and telemetry actions, and describes how to use it as the FROM image to install components and build your own agent runtime image.
Built on Alibaba Cloud Linux 4, this image provides the Agentic OS component installation CLI (anolisa) and the component repository configuration, but does not preinstall the binary of any Agentic OS component. This allows you to install components as needed and control the image size and attack surface on your own.
Image versions
|
Image name |
Address |
Information |
|
Agentic OS base image |
|
anolisa-cli: 0.3.8 |
All examples in this topic use the 1.3 tag. Replace the tag accordingly when you upgrade to another version.
Suitable scenarios
Use this image as the FROM base image, install components such as AgentSight, SkillFS, ws-ckpt, and AgentSecCore as needed, and build your own agent runtime image. You can also use it as an interactive environment to try out components and troubleshoot issues.
If you need an out-of-the-box runtime environment for a single component, use the corresponding component image instead. This image contains no foreground daemon process and cannot be used directly as a long-running service image.
Quick start
# Enter the interactive environment
docker run -it --rm ac2-registry.cn-hangzhou.cr.aliyuncs.com/ac2/agentic-os:1.3
# Check the CLI version and available subcommands
docker run --rm ac2-registry.cn-hangzhou.cr.aliyuncs.com/ac2/agentic-os:1.3 anolisa --version
This image has no long-running foreground process. If you do not specify-itor a command, the container exits immediately with code 0. In Kubernetes, the container is displayed asCompletedand is restarted repeatedly. To debug a pod, explicitly passargs: ["sleep", "infinity"].
Image information
|
Item |
Value |
|
Image address |
|
|
Current default tag |
|
|
Naming convention |
The |
|
Architecture |
|
|
Base image |
Alibaba Cloud Linux 4.0.5 |
|
Component installation CLI |
|
|
Runtime user / working directory |
|
|
Listening port / built-in environment variables |
None (neither |
|
ENTRYPOINT / CMD |
|
On each startup, the entrypoint first runs anolisa telemetry enable (Internet access is required. If the operation fails, only a Warning: message is printed and the startup is not blocked), and then runs exec on the command that you pass in. If no command is passed in, the container enters bash. To skip this step, override the entrypoint by using --entrypoint (Docker) or command (Kubernetes).
Use as a base image for builds
The component repository alinux-agentic-os-repos is already configured. You can install components directly after FROM. The following example uses AgentSight:
FROM ac2-registry.cn-hangzhou.cr.aliyuncs.com/ac2/agentic-os:1.3
# The default backend of the CLI is raw. Switch the backend before you install components by using the rpm backend.
RUN sed -i 's/default_backend = "raw"/default_backend = "rpm"/' /etc/anolisa/repo.toml && \
yum makecache && \
yum install -y agentsight && \
yum clean all && rm -rf /var/cache/dnf/*
# Your foreground service. If you do not specify it, the entrypoint of the base image is used and bash is started.
CMD ["agentsight", "--help"]
The installation methods of the components are described in the following table.
|
Component |
Installation command |
|
AgentSight |
|
|
SkillFS |
|
|
ws-ckpt |
|
|
AgentSecCore |
|
Take note of the following three points:
-
The runtime dependencies of a component are not automatically installed. Install the required packages as described in the documentation of the corresponding component. Otherwise, a
command not founderror is reported only at runtime. -
yum clean all && rm -rf /var/cache/dnf/*reduces the image size only when it is specified in the sameRUNinstruction asyum install. -
Component versions are independent of the image tag. If you require reproducible builds, specify a fixed component version in the command.
Image update policy
This image is released in line with the release cadence of Agentic OS, and the interval between releases does not exceed one month. After a new version is released, the previous version is no longer maintained. Issue fixes and component upgrades are applied only to the latest version. We recommend that you upgrade at the earliest opportunity.
For builds and deployments, use a tag that contains a version number, such as 1.3, instead of relying on latest. Otherwise, a rebuild may silently switch to another version.