Cloudpods: Open-Source Unified Management for Multi-Cloud, On-Premise, and AI Workloads

What Cloudpods Solves and Who Needs It

Cloudpods addresses the operational fragmentation that results from managing multiple cloud providers and on-premise infrastructure through separate tools and consoles. Each public cloud has its own SDK, its own authentication model, its own resource naming conventions, and its own pricing. On-premise hypervisors like KVM and VMware have their own management layers that are completely separate from public cloud control planes. Cloudpods presents a single API over all of these.

The README describes it as a cloud on clouds: a platform that runs on top of existing cloud and on-premise infrastructure rather than replacing it. The intended users are teams that need a unified view of both public and private cloud resources, teams managing multiple cloud accounts from different providers and wanting centralized access control, teams virtualizing a small number of physical servers into a private cloud, teams managing VMware clusters who want to expose those resources through a cloud-style API, and teams deploying LLM inference services and AI containers who need GPU scheduling and workload management. The README frames the platform as suitable for organizations anywhere from a few physical servers to a large hybrid cloud deployment.

Architecture: A Go Platform Wrapping Heterogeneous Infrastructure

Cloudpods is written entirely in Go. The module path is yunion.io/x/onecloud, reflecting the project's earlier internal name. The go.mod file shows dependencies on cloud-provider SDKs for AWS (aws-sdk-go-v2), Alibaba Cloud, and others alongside infrastructure management libraries for VMware, IPMI/Redfish, and container runtimes including containerd.

The repository is organized into cmd/ for the command-line entry points, pkg/ for the core Go packages, vendor/ for vendored dependencies, and scripts/ for deployment tooling. The build system is Make with a central Makefile that sets linker flags to embed git version, commit, and branch information into the binary. The build script at build/build.sh handles cross-platform compilation.

The platform exposes a unified REST API and a Swagger API documented at cloudpods.org/en/docs/swagger/. It hides the differences between underlying cloud providers and exposes one consistent resource model. A virtual machine on AWS, a KVM instance on on-premise hardware, and a Proxmox VM are all exposed through the same Servers API endpoint.

Supported Cloud Providers and On-Premise Resources

The public cloud provider list in the README is extensive. Supported providers include AWS, Azure, Google Cloud Platform, Alibaba Cloud, Huawei Cloud, Tencent Cloud, UCloud, Ctyun (China Telecom), ECloud (China Mobile), JDCloud, Volcengine, Baidu Cloud, Kingsoft Cloud, QingCloud, Oracle Cloud, and China Unicom Cloud.

For private cloud and virtualization, Cloudpods supports OpenStack, ZStack, Alibaba Cloud Apsara, Huawei HCSO, Nutanix, BingoCloud, and OceanBase.

For on-premise resources, it supports a lightweight private cloud built on KVM, VMware vSphere vCenter and ESXi, Proxmox VE, baremetals managed through IPMI and Redfish API, object storage via Minio, Ceph, and XSky, and NAS via Ceph.

The resource types managed through the unified API include servers, disks, network interfaces, networks, VPCs, storages, hosts, snapshots, security groups, elastic IPs, load balancers (instances, listeners, backend groups, certificates, ACLs), object storage buckets and objects, NAS file systems and mount targets, RDS instances, elastic cache, DNS zones and records, and VPC peering and NAT configurations.

AI Cloud: LLM Inference and GPU Workload Management

The README describes an AI Cloud feature that adds a unified management layer for large language model inference and AI container applications. It is presented as integrated with the broader Cloudpods resource ecosystem rather than a separate product.

The AI inference component supports deploying and managing LLM inference instances using Ollama, vLLM, and SGLang as runtime backends. It handles GPU scheduling, model mounting, and inference service address allocation. The model library component provides unified management of model sources, versions, and caches with support for multi-instance reuse and offline distribution.

The AI application management layer supports one-stop deployment of LLM application orchestration tools, agent assistants, image generation applications, and other AI container workloads. Specifically named supported applications include OpenClaw, Dify, ComfyUI, and Hermes Agent.

GPU operations are handled through automatic device detection and registration, with unified NVIDIA and CUDA environment configuration managed by the platform. Templates define resource specifications including CPU, memory, and GPU counts, while images manage the container runtime environments for AI workloads.

Getting Started and the API Surface

The README points to cloudpods.org/en/docs/getting-started/ for installation documentation, organized by deployment scenario. The Makefile at the repository root provides the standard Go build workflow:

make

This compiles all binaries using the build script and flags set in the Makefile. The Makefile supports parallel builds using the -j flag. Version information including git commit, branch, and build date is embedded into the binaries at compile time through ldflags.

The Swagger API documentation at cloudpods.org/en/docs/swagger/ documents the full resource API. The unified API is the primary way to interact with all managed resources: provisioning, lifecycle management, monitoring, and policy enforcement all go through the same endpoint structure regardless of the underlying provider.

For operators managing the platform itself, a CLI tool is available. The Cloudpods documentation at cloudpods.org/en covers installation guides, configuration references, and architecture documentation.

Comparison with OpenStack and Licensing

OpenStack is an open-source platform for building private clouds, primarily targeting on-premise infrastructure. It provides a comprehensive set of services for compute, networking, storage, and identity management within a single data center or collection of on-premise servers. OpenStack can be connected to public cloud providers through additional components, but its core design is oriented toward managing infrastructure you own rather than presenting a unified view across accounts you access via third-party APIs.

Cloudpods takes the opposite starting point: it treats both on-premise and public cloud resources as equal first-class targets and builds the unified API over all of them from the start. For a team that already has heavy OpenStack investment in their on-premise environment, adding Cloudpods creates a second management plane rather than consolidating one. For a team starting fresh that needs to manage both on-premise and multiple public cloud accounts, Cloudpods provides that unified view without requiring them to run a full OpenStack deployment first.

Cloudpods is licensed under Apache-2.0. There are no commercial restrictions on use, modification, or redistribution, provided the license file is included and copyright notices are preserved. The module path yunion.io/x/onecloud reflects the project's earlier internal code name; external package imports should use that path rather than the GitHub repository URL.

Editorial conclusion

Cloudpods is suited for infrastructure teams that manage a combination of on-premise servers and multiple public cloud accounts and want a single control plane across all of them. The repository received a push on 2026-09-27 and is under active development. Teams that work exclusively with a single public cloud provider, or that only manage on-premise resources without any multi-cloud requirement, should evaluate whether the operational overhead of a platform this broad is justified for their scope. The Apache-2.0 license imposes no commercial restrictions.

Frequently asked questions

Official sources

1. Issues
2. License: Apache-2.0
3. Project website
4. README
5. yunionio/cloudpods on GitHub