Best Nextcloud Apps in 2026: What's a Must-Have and What to Avoid for Real Productivity

Nextcloud Hub 26 Winter, released in February 2026, runs on over 500,000 servers worldwide and ships with more than 400 apps in its app store. That number is both its greatest strength and its biggest trap. Install the right apps on a well-tuned VPS and you have a private, self-hosted alternative to Google Workspace that outperforms the competition. Install the wrong ones and you will spend your days chasing slow load times, runaway cron jobs, and database tables nobody asked for. This guide cuts through the noise: here is what to install, what to skip, and why — based on real resource impact and actual use cases.

The Foundation: Core Apps Worth Keeping Enabled

These apps ship bundled with Nextcloud and are enabled by default. Most operators leave the defaults untouched and move on. That is a mistake — not because the defaults are wrong, but because understanding what each one does lets you make deliberate choices about what your instance actually needs to support.

Nextcloud Files is the non-negotiable core. Everything else depends on it. Hub 26 Winter introduced a complete rewrite of the file access layer — the new ADA (Accelerated Direct Access) engine computes permissions in advance, makes the mount table authoritative, and enables direct S3 downloads for large instances. If you are on Hub 26, you are already benefiting from this without any configuration. Keep Files enabled and keep it updated.

Nextcloud Talk replaces Zoom, Teams, and Slack in a single app. Hub 26 adds pinned messages, scheduled messages, live AI transcription with translation, hardware-accelerated background blur, and auto-gain/echo cancellation for calls. For any team already on Nextcloud, disabling Talk to use an external chat tool is a lateral move that only fragments your data footprint. The one caveat: for calls with more than a handful of concurrent participants, you need the High Performance Backend (HPB) running as a separate service — without it, call quality degrades sharply above 4–5 simultaneous users.

Nextcloud Office (powered by Collabora Online) is the answer to Google Docs. Hub 26 adds document comparison (side-by-side and tracked changes), Sheet View for independent filtering in shared spreadsheets, synchronized slide presentations ("Present to All"), and LaTeX support in Nextcloud Text. For teams that collaborate on documents, this is a must. For a solo user who only needs file sync, you can disable it and reclaim the memory that the Collabora container consumes — typically 300–500MB RAM at idle.

Must-Have Apps Beyond the Defaults

Nextcloud Deck is Trello done right, self-hosted. Hub 26 adds federation support — you can now share boards across different Nextcloud instances — and board embedding in Whiteboards. For project management without SaaS subscriptions, Deck is lightweight, well-maintained, and integrates natively with Calendar and Talk. It is the first app to install after the core suite.

Nextcloud Collectives fills the knowledge base role. Think Notion or Confluence, but stored as plain Markdown files you own. It is built on top of Nextcloud Text and integrates directly with the file system, which means your documentation is searchable, versionable, and exportable without vendor lock-in. Hub 26 improved its resilience under unstable connections — a real improvement for remote teams on variable-quality internet. If you run a team and have no internal wiki, Collectives belongs on your instance.

Nextcloud Tables covers the structured data use case that neither Files nor Deck handles well — think lightweight databases, CRM lists, inventory sheets, or editorial calendars. Hub 26 adds advanced date filters and public read-only share links. It is not a replacement for a proper relational database, but for non-technical users who would otherwise build spreadsheets in Google Sheets and email them around, Tables eliminates that entire workflow.

Two-Factor Authentication (TOTP) and Brute Force Protection are not glamorous, but on a publicly accessible instance they are mandatory. Both are available in the app store. Brute Force Protection is already included in Nextcloud core from version 12 onward, but verify it is active. TOTP adds a second authentication factor that stops credential-stuffing attacks cold. If your Nextcloud is reachable from the public internet without a VPN in front of it, these two apps are the difference between a secure server and an incident report.

Nextcloud Whiteboard got a significant upgrade in Hub 26 — timers, voting, reactions, comments, table embedding, and version history. For teams that run workshops, planning sessions, or retrospectives, it replaces Miro or FigJam completely. It does require a separate Whiteboard server component to run, which adds operational complexity, but for teams that use it regularly the setup cost pays off quickly.

Apps to Avoid (or Treat with Caution)

Full Text Search (Elasticsearch) is the most resource-intensive optional component in the Nextcloud ecosystem. It requires a running Elasticsearch or OpenSearch instance alongside your Nextcloud server, which means a separate process consuming 1–2GB of RAM before indexing a single file. On a VPS with 4GB of RAM running a full Nextcloud stack, this leaves your PHP-FPM workers starved and your database competing for memory. Unless your instance has hundreds of thousands of files and users who actively search document content, the native Nextcloud search is sufficient and costs nothing. Install Full Text Search only if you have dedicated resources for it.

Recognize is the AI-powered photo and audio recognition app that automatically tags faces, objects, and scenes in your media library. The concept is appealing. The reality on a modest VPS is that the background indexing job will saturate your CPU for hours after uploading a photo album, degrade response times across the entire instance, and fill your Nextcloud log with processing entries. On a dedicated server with spare CPU cores it works as advertised. On a shared VPS or a small instance handling other workloads simultaneously, it is a performance liability. Disable it unless you have benchmarked your server under its load.

Suspicious Login uses a machine learning model to flag unusual login patterns. The intention is sound. The implementation trains and runs a local ML model against your login history, which creates background CPU and I/O load that grows with your user count. For instances with fewer than 50 users, the threat model it addresses is better handled by TOTP and Brute Force Protection at a fraction of the resource cost. The two-factor authentication approach stops the attack at the front door; Suspicious Login tries to detect it after the fact, at ongoing server cost.

The Social app was Nextcloud's attempt at a federated social network via ActivityPub. It has seen minimal development, poor adoption, and a long tail of open issues. As of Hub 26 it remains in the app store but is effectively stagnant. Installing it adds database tables, background jobs, and federation endpoints for a feature set that adds nothing to a productivity-focused instance. Skip it entirely.

Preview generation for all file types is not a single app but a configuration trap. By default, Nextcloud generates previews for images, PDFs, and office documents. Enabling preview generation for video files (via the Preview Generator app combined with ffmpeg) will trigger a background job that transcodes every video file you upload — a process that can consume 100% of a CPU core indefinitely on an instance with an active media library. If you store video files on Nextcloud, either disable video preview generation or run it explicitly on a schedule during off-peak hours with resource limits applied via cgroups.

AI Features: Use Selectively

Hub 26 significantly improved the performance of the Nextcloud Assistant — up to 10 seconds faster per prompt, 5–8x faster local transcription. It also added Memories (persistent context across AI conversations), AI content labeling in compliance with the EU AI Act, and support for Olmo 3 and self-hosted IBM watsonx. The key word is self-hosted: the privacy case for Nextcloud's AI features only holds if you run the models locally. Connecting Nextcloud Assistant to OpenAI's API means your data leaves your server and enters OpenAI's infrastructure, which defeats the entire data sovereignty argument for running Nextcloud in the first place. If you use AI features, use locally running models via Ollama or the bundled LLM2 backend. If you connect to external APIs, understand the trade-off explicitly.

By Use Case: The Short Decision Guide

Use case	Install	Skip
Solo user, file sync only	Files, TOTP, Brute Force Protection	Office, Talk HPB, Recognize, Full Text Search
Small team (2–15 people)	Files, Talk, Office, Deck, Collectives, TOTP	Recognize, Social, Suspicious Login
Project-heavy team	All of the above + Tables, Whiteboard	Full Text Search unless >100k files
Media-heavy personal instance	Files, Photos, TOTP	Recognize (unless dedicated server), video previews
Developer / power user	Files, Collectives, Tables, Assistant (local)	Social, Suspicious Login, unnecessary groupware

The Server Side of the Equation

Every app recommendation above comes with an implicit assumption: that your server is sized appropriately for what you are running. The most common reason a Nextcloud instance feels slow is not a misconfigured app — it is an undersized VPS. A minimal Nextcloud instance serving one to five users needs at least 2GB of RAM, a PHP-FPM pool tuned to your worker count, Redis for memory caching and session locking, and a database with a properly sized buffer pool. Add Collabora for office editing and that baseline jumps to 4GB. Add Whiteboard's separate server component, the High Performance Backend for Talk, and local AI models via Ollama, and you are looking at 8GB minimum to avoid constant memory pressure.

Running your own Nextcloud on a VPS gives you complete control over which apps run, what resources they consume, and where your data is stored — none of which is true on a managed Nextcloud hosting plan. A KVM VPS with dedicated resources, a clean Debian or Ubuntu base, and Nextcloud installed via the official zip or AIO Docker image is the most reliable setup in 2026. If you want the full self-hosted Nextcloud experience without hardware management, a dedicated KVM virtual server with guaranteed RAM and NVMe storage gives you the foundation to run a production-grade instance for a team of any size.