Today, we are releasing the latest updates to the EdgeRunner platform. This release focuses on delivering greater control, improved performance, and a more streamlined overall experience.

With the introduction of a new modular installer and model delivery framework, users can more easily install, manage, and update models and system components based on their needs. At the same time, major performance improvements—from 20% faster document ingestion and reduced memory usage to 2x faster model prefill and up to 7x compression gains—make core workflows faster and more efficient. 

Complementing these updates are key usability enhancements, including full web application support for seamless access without local installation, deep-link citations that allow users to jump directly to highlighted source content for verification, and Magazine (.mag) file support for portable, secure transfer of chats, documents, and vector data across environments—making the platform more flexible and practical across real-world use cases.

What’s New

• Full Web Application Support – The entire application is now fully supported in a web environment. Users can access the platform without local installation, enabling easier onboarding and hosted use cases while maintaining feature parity with the desktop experience.
• Deep-Link Citations – When performing Retrieval-Augmented Generation (RAG), users can click an “Open Page” button that takes them directly to the exact location in the source document. The retrieved section is automatically highlighted, making it easier to verify context and trace answers back to their origin.

• Magazine File Support – Introducing Magazine files (.mag), a portable archive format for exporting and importing chat histories, documents, and collections as a single compact file—making it easy to transfer, share, or back up work without losing context. With support for EdgeRunner Magazines, this allows secure, near-instant migration of databases and vector stores across environments, with built-in encryption and signature support to ensure trusted and secure file transfer.

• Smart Rich Copy + Markdown Copy – Copying content now preserves formatting in tools like Word and Excel. Additionally, a new “Copy as Markdown” option enables clean pasting into plain-text or developer environments.
• LaTeX Math Support – Full LaTeX support has been added, with clean rendering in chat and preservation across all export formats for technical and analytical use cases.
• In-Chat Search (Cmd/Ctrl + F) – Users can now search within conversations using familiar shortcuts, making it easy to locate specific content.

• Expanded Export Formats – Chats can now be exported as Word, Markdown, plain text, or Magazine (.mag), providing flexibility for different workflows and sharing needs.
• Collapsible Sidebar – A collapsible sidebar allows users to focus on their workspace, especially when working with long-form content.

• Custom Rendering Engine – Introduces a custom linear-time rendering system for large, stylized messages, significantly improving performance and responsiveness in complex outputs.

Installer Updates

• New Modular Installer + Updater – We’ve introduced a brand-new installer that gives users full control over what gets installed. Users can choose specific components such as models and runners, enabling a more tailored setup based on their hardware constraints, storage, and mission requirements. This system also acts as an updater—allowing users to seamlessly install new models, update existing ones, and manage their environment.

• Model Delivery Framework – Models are now modular, versioned components that can be distributed and updated independently. This enables faster rollout, reduced overhead, and greater flexibility across environments.
• Faster Installation & Delivery – Leveraging the new delivery framework, installation speeds are now >2x faster (depending on network conditions), with optimized pipelines that reduce download overhead and improve reliability.
• Dynamic Distribution Foundation – Establishes the groundwork for environment-specific deployments and future dynamic provisioning of models and runtime components.

Other Improvements

• Performance Upgrades – Optimizations across the platform deliver faster processing, improved accuracy, and reduced resource usage across key workflows.
  • Document ingestion is now 20% faster with reduced runtime memory requirements, alongside overall system memory optimizations for more efficient performance.
  • End-to-end RAG pipeline accuracy has been improved, resulting in more reliable and contextually precise responses.
  • Model performance has been enhanced with 2x faster prefill and 10% faster token generation on the Medium model, enabling quicker response times and smoother interactions.
  • EdgeRunner Compression introduces up to 7x compression and speed improvements while maintaining parity in MOS-specific workloads, along with a 50% reduction in KLD and 50–90% error reduction in evals compared to alternative methods.
• Stability & General Enhancements – Various minor issues have been resolved alongside additional UI refinements and workflow improvements, resulting in a more reliable and polished user experience.

To get started with the latest version of EdgeRunner, click here.