openEuler's Latest Releases Embed AI Agents and Developer Tools Into the OS
Two new versions of the Linux distribution—24.03 LTS SP4 and 26.09—bring AI-powered diagnosis, self-hosted agent platforms and user-space package management to developers and enterprises.
The openEuler project has shipped two new releases: version 24.03 LTS SP4 in July 2026, followed by the shorter-cycle 26.09 innovation release on October 1. The long-term support track continues to receive feature updates through service packs, while the innovation releases cycle through support on a faster schedule.
openEuler maintains two distinct release channels: long-term support versions that receive updates through numbered service packs, and innovation releases that introduce newer technologies on shorter support windows. The 24.03 LTS line will continue receiving updates, with the next major LTS version anticipated in 2028. Innovation releases like 26.09 are supported until March 2027, mirroring the cadence of distributions like Ubuntu's six-month cycle.
Advances in 24.03 LTS SP4
The latest service pack remains on Linux kernel 6.6 but incorporates backported features from newer upstream kernel releases.
AI-driven system analysis and optimization
The most immediately useful AI capability for typical users involves automated system diagnosis and performance tuning. An Intelligent Diagnosis Agent collaborates with a Known Issue Analysis Agent to identify problems, determine underlying causes and propose solutions. The tuning component gathers performance metrics, identifies bottlenecks and delivers optimization recommendations—potentially eliminating hours spent analyzing system monitoring tools.
These capabilities ship through openEuler's Witty framework, installable via the witty package from standard repositories. Users launch the interface with witty or opencode and communicate with diagnosis and tuning agents using natural language. Unlike earlier iterations of EulerCopilot that relied on openEuler's proprietary cloud AI infrastructure, Witty accommodates both locally-running models and OpenAI-compatible APIs, expanding accessibility for deployments outside China.
The release also introduces a lightweight Agent sandbox that isolates code execution and tool invocations from AI agents, a practical security measure given the prevalence of granting AI systems shell access.
Kernel-level improvements
Several under-the-hood changes enhance system capabilities. The sched_ext framework, an eBPF-based scheduler that arrived in Linux 6.12, allows custom CPU schedulers to load at runtime without kernel recompilation. If a custom scheduler fails, the kernel automatically reverts to its default scheduler.
FUSE file systems now communicate with the kernel via io_uring, potentially improving performance for FUSE-based utilities like sshfs. A programmable page cache enables applications to define their own file caching strategies, designed to accelerate loading of large AI models where disk I/O typically becomes the limiting factor.
ARM-based systems gain the option to select a 64K page-size kernel during installation, though the 4K kernel remains the standard default.
Enterprise and data center features
SP4 introduces NPU compute slicing, accelerated recovery for AI inference containers, SuperPoD networking and reliability enhancements, and confidential computing improvements for Arm CCA, TaiShan and Hygon systems. openEuler's own compiler builds—LLVM, Go and BiSheng JDK—received updates as well.
What's included in 26.09
The 26.09 release ships with two kernel versions: Linux 6.6 for server deployments and Linux 6.18 for DevStation, openEuler's developer-focused desktop edition. Most desktop-oriented additions concentrate in DevStation.
Installation and package management
DevStation introduces a new installer called heolleo, which uses SquashFS-based images, supports offline installation from local ISO files and requires no network connectivity.
A user-space package manager called epkg allows software installation in home directories without modifying system packages. Users can maintain isolated environments for separate projects and configure multiple package sources, resembling tools like Conda or Nix.
Self-hosted AI infrastructure
DevStation bundles PolyMind, a self-hosted platform for AI chat, agent workflows and multi-model management. The platform integrates with MCP services and AI agents through its agentd service.
openEuler launched SkillHub, a repository for discovering, installing and distributing AI agent skills. Each skill undergoes automated security assessment, though openEuler acknowledges this aids risk evaluation without guaranteeing absolute safety.
AgentCensor operates between AI agents and the system, recording tool calls, file modifications and system calls while allowing, blocking or requesting approval before agent execution. SCCS addresses token bloat in long-running agents by replacing large tool outputs in the context window with compact references, retrieving full content only when necessary. Additional tools—ram-a-kv, MemFabric and ModelFS—optimize inference server performance.
ModelFS accelerates model startup by leveraging the programmable page-cache framework to tune caching behavior for AI workloads while maintaining compatibility with existing applications.
Downloading openEuler
Both releases are available from the official download page, with variants for servers, embedded devices and desktop systems. DevStation is the desktop edition and includes the AI tools described above.
The distribution has evolved substantially since its previous coverage two years ago. The emphasis on AI integration reflects the current technology landscape in 2026. An operating system targeting developers must now provide AI capabilities as a core component.