Nextcloud introduces ADA engine for massive performance improvements

Contents

With the ADA engine (Accelerated Direct Access), Nextcloud has introduced a fundamentally revised data access architecture. The engine, implemented in PHP, Go, and Rust, is intended to elevate the scalability of the free collaboration platform to a new level. The new development is named as a tribute to Ada Lovelace, the first computer programmer in history.

The ADA engine pre-calculates access data and permissions, stores them in the cache, and enables direct file access. It also actively pushes data to clients to make navigation more responsive.

Concrete performance leaps in Nextcloud Hub 26 Winter

Users can experience the new ADA engine in action for the first time with Nextcloud Hub 26 Winter, which is scheduled for release on February 18, 2026. The new version separates previews from the file cache, thus reducing the size of the file cache table by 56 percent. This metadata abstraction layer is often the largest database table in Nextcloud installations. Previews will receive their own table with an expiration mechanism for unused files.

Authoritative Mount Points accelerate the loading of folders with shares by 30 percent, in Nextcloud's example, from 1.9 to 1.3 seconds. The Lean File System Setup improves shared folder retrieval by 60 percent, from 1.39 to 0.44 seconds. Direct S3 downloads massively reduce server load and accelerate thumbnail loading by a factor of 2 to 10.

Snowflake IDs and Sharding for large installations

A central innovation is Snowflake IDs, originally developed by Twitter. These 64-bit identifiers can be generated decentrally without database queries and contain a timestamp with millisecond precision, a server ID, and a CLI flag. They enable sharding, i.e., splitting tables by user or file IDs across multiple nodes, which reduces waiting times. Snowflake IDs are already in use in the preview provider and external sharing.

For installations with millions of users, Nextcloud also implements the generator pattern for streaming large lists. Instead of loading complete lists into memory, which can lead to out-of-memory errors, the data is processed step by step. A new mount points table replaces the previous per-user caches and enables direct provider queries. The architecture is ideally suited for clustered and cloud-native deployments such as Kubernetes.

High-performance backends for Files and Talk

The high-performance backends developed in Rust and Go also receive updates. HPB Files 2.0 reduces PROPFIND requests for updates by 80 percent through staggered notifications for multi-user changes and more detailed information for selective sync. HPB Talk 2.0 introduces a chat relay and reduces database load for large rooms and calls. With more than 100 participants, chat-related requests decrease by up to 80 percent.

Direct S3 downloads are already implemented in the updated desktop client. Clients load directly from S3-compatible storage in a token-protected manner, bypassing the application server. Corresponding functions for the web interface and previews will follow in later releases. The storage abstraction remains in place and continues to support POSIX, S3, IBM, FTP, WebDAV, Samba, NFS, and SharePoint.

Security and open-source character remain intact

Despite the profound changes, all security features remain or are improved: server-side and end-to-end encryption, ACLs, password and expiration mechanisms, and video verification. AI-powered detection of suspicious logins, brute-force protection, rate limiting, audit logging, detection of sensitive files, and smart locking also continue to function. The ADA engine ensures consistent permissions across all features, from Files to Talk to Tasks. Detailed information on all changes can be found in the Nextcloud blog.

And of course, the open-source character of Nextcloud remains untouched; this is not a commercial pro feature. The first beta of Nextcloud Hub 26 Winter has been available since January 2026. Nextcloud promises that further adjustments will follow in upcoming releases, with the greatest impact expected in large installations. Geographically distributed storage solutions are technically feasible with ADA but still require significant development effort.

(fo)