(I have no inside knowledge, but I'd be quite surprised if they implemented any AI-heavy features like 'chat with PDF', especially given the wide availability of plugins with all sorts of different features and approaches)

My reading is that the plugin route is best for niche features that benefit specific use cases. However, I would argue that artificial intelligence belongs at the heart of Zotero's future: as a core utility for managing, organizing, and synthesizing research, rather than an optional add-on.

Relegating AI entirely to third-party plugins risks fragmenting the user experience. Modern academic workflows rely increasingly on semantic search, automated tagging, smart summarization, and discoverability: capabilities that should be deeply integrated into Zotero’s native database architecture rather than patched on top. By building privacy-first AI tools directly into the application, Zotero can ensure these features remain open-source, locally processed, and seamlessly aligned with its core mission.

While excellent plugins like Beaver and PapersGPT get the job done for AI-literate power users, a vast segment of the Zotero community doesn't yet realize how transformational a well-integrated, native AI workflow could be for their daily research.

@adamsmith I appreciate that perspective. Still, relying solely on third-party plugins for AI features risks turning Zotero into the "Nokia of reference managers", failing to adapt natively until a more modern competitor captures the user base and falling into irrelevancy.

Here is why native integration is vital, even in a polarizing landscape for AI:

1. Solving the Polarization & Privacy Dilemma:
Because AI is polarizing, native integration actually gives Zotero the power to set the rules. By building it into the core app, Zotero could offer a locally run, on-device model by default. This directly addresses the privacy and data-harvesting concerns that make some users hesitant about third-party AI tools.

2. The Risk of Fragmentation:
While plugins can do almost anything, the current AI plugin ecosystem is incredibly fragmented. Forcing users to audit multiple third-party tools just to find one with a trustworthy privacy policy creates a massive barrier to entry.

3. AI as a Core Competency, Not an Add-on:
Semantic search, automated metadata extraction, and contextual synthesis are quickly becoming baseline expectations for research tools, not "advanced features." Leaving this to the plugin ecosystem means Zotero's core experience risks falling behind newer, AI-native research tools.

So to go on your point, native implementation doesn't mean forcing AI on everyone. It means offering a secure, privacy-first, opt-in baseline that protects users, while leaving the door open for plugins to expand on it for niche features.

@raphael.baril maybe I am mistaken, but your last post sounds like Claude is trying to develop your arguments from your previous post, in generic way (sorry if it is not the case; I see too often these common AI language patterns to not become too suspicious. ); AI is important, but done in bad way is just slop;
Also, the methods of using AI bots with external programs are changing (now it is MPC/skills, but this is evolving; who now what protocol will be cool next year? Plugins & external tools give enough flexibility, and Zotero ecosystem is open enough that is super easy to plug external tools; recently developers of DevonThink (let's say an analogous app for work with text) added mcp and AI there but the app is a proprietary tool where users cannot build their own tools). Adding local tools models is nice, but for many use cases, it might be not enough/not everyone will have enough

I agree with @adamsmith that the topic is polarizing; if one of the most core features, citation management for text based workflows (LaTeX/Markdown), is managed by a plugin, I don't see a reason why the ai features cannot by handled by plugins or external tools/libraries (i someone want to read how to use MCP with Zotero, my write-up is here https://danielborek.me/2026/zotero-mcp-ai/)

Also AI is often just subtly or less subtly wrong, while it is the individual researcher's responsibility to check, & Zotero team endorsing it may be too much for some of the folks; and lastly, arguments, sites like the one posted in the first post, are for profit & selling tokens; this is not free gift for us researchers :)

The last point - every use of AI is niche in that sense that there is no AI workflow that fit to all, that why there is so many plugins; also if developers think better how to highlight better plugin (Zotero might benefit more from something like official marketplace for plugins, which review/user ratings, maybe a bit like Obsidian, whose developers also didnt choose to develop AI support themselves, instead left it to community)