News
What happened
With the release of Dragonfly v2.5.0, you can now directly download model repositories from Hugging Face and ModelScope, streamlining your workflow. This update also introduces new features like comprehensive rate limiting and a blocklist for download control, which enhance your deployment's stability and security.
Dragonfly v2.5.0 has been released, introducing several new features and enhancements that are particularly beneficial for your self-hosted infrastructure. You can now directly download model repositories from Hugging Face and ModelScope, making it easier to access machine learning models. Additionally, the new Dragonfly Injector for Kubernetes allows for automatic P2P capability injection into application Pods, enhancing your deployment's efficiency. The update also includes improved rate limiting and download control features, ensuring better management of your resources under high load.
Release at a glance
Key facts from the announcement.
Version
2.5.0
Product
Dragonfly
Released
June 30, 2026
Platform
Kubernetes
PRIVACY STACK
Extend Privacy Beyond DNS
Controlling your DNS queries is one layer of network privacy. Your email metadata — who you talk to, when, how often — is equally exposed with standard providers. Proton Mail applies end-to-end encryption to the layer most people ignore.
Try Proton Mail →This is an affiliate link. If you purchase, I earn a commission at no extra cost to you.
Changes at a glance
What's new
The Dragonfly Client now supports downloading model repositories directly from Hugging Face and ModelScope, which simplifies your model management process. The introduction of the dfctl command-line tool allows you to manage local storage tasks more effectively, including listing and removing resources.
Furthermore, the new blocklist feature lets you control downloads to prevent service disruptions, while enhanced rate limiting capabilities across the control plane and client help maintain system performance under load.
Breaking changes
No breaking changes were reported in the source material.
Analysis
In detail
Dragonfly v2.5.0 introduces direct repository downloads from Hugging Face and ModelScope. You can now run commands like `dfget hf://deepseek-ai/DeepSeek-OCR` to fetch repositories directly. Git LFS data is downloaded using Dragonfly's P2P acceleration, while other metadata is fetched through the Git protocol.
The new Dragonfly Injector for Kubernetes allows you to inject Dragonfly client binaries and configurations into application Pods without needing to rebuild container images. This feature is enabled through annotation-based policies and is supported by Helm Charts for easy deployment.
Additionally, Dragonfly now includes a blocklist feature in the Manager console to disable specific downloads, which can help mitigate the impact of abnormal requests. The comprehensive rate limiting capabilities introduced in this version allow you to configure request limits for various operations, enhancing your system's stability during high load.
Key takeaways
The most important facts from this update.
Why it matters
These enhancements in Dragonfly v2.5.0 significantly improve your ability to manage machine learning models and optimize resource usage in your homelab. The new features help ensure that your deployments are more stable and secure, especially under high load conditions.
Homelab impact
With the ability to directly download model repositories, you can streamline your workflow and reduce the time spent managing model files. The automatic P2P capability injection simplifies your Kubernetes deployments, allowing you to leverage Dragonfly without extensive configuration changes.
The introduction of rate limiting and download control features provides you with better tools to manage resource usage and protect your services from unexpected spikes in demand. This means you can maintain a more resilient and efficient homelab environment.
What to do next
Practical steps for operators running self-hosted stacks.
This brief covers what you need from CNCF Blog's reporting. Visit the original post for release notes, changelogs, and full technical documentation.