Rethinking my PKM: Moving beyond Evernote

TL;DR: After more than a decade with Evernote, I’m evaluating a move because my archive has grown to about 42 GB across 6,311 files, and I want better long-term control over the underlying data. The two strongest replacements are Obsidian and DEVONthink. Obsidian has the architectural advantage: notes and attachments live as ordinary files, which minimizes lock-in and makes the archive especially well suited to future AI tools. DEVONthink has the operational advantage: it is stronger at managing large collections of PDFs, scans, images, and mixed documents, has built-in AI/MCP integration, and is likely cheaper over ten years. I do not want to split my knowledge across multiple apps, so the goal is one canonical repository. The next step is a controlled migration pilot using the same representative Evernote subset in both systems, testing fidelity, sync, mobile use, document handling, AI retrieval, and ease of eventually leaving the new platform.

After more than a decade as a paying Evernote user, I am considering moving my personal knowledge archive to a different platform. Evernote has served me well, but the accumulated size and value of the archive have made one concern increasingly important: I do not want decades of personal information locked inside a proprietary application and data model.

This is no longer simply a note-taking application decision. My Evernote account now contains approximately 42 GB and 6,311 files, including thousands of text notes, hundreds of PDFs, images, scans, and other attachments. I routinely read and write this material from a Mac laptop, iPhone, and iPad. Whatever replaces Evernote must function as a unified personal information repository across all three devices.

Generative AI has also changed the requirements. A modern personal knowledge system should make it possible for an AI assistant to search and reason across the entire corpus—for example, to answer a question using relevant notes, PDFs, images, receipts, research, or other material accumulated over many years, while ideally citing the underlying sources.

After considering a broader set of alternatives, Obsidian and DEVONthink are the two leading candidates. My current inclination is that the final decision should be made after a controlled migration pilot rather than from feature comparisons alone.

The central tradeoff is straightforward:

Obsidian offers the stronger long-term data architecture. DEVONthink offers the stronger integrated document-management application.

Why Leave Evernote?

The problem is not that Evernote has stopped being useful. In fact, it continues to add significant functionality, including AI features, semantic search, and connectivity to external AI tools through MCP. Its current Advanced plan is designed specifically for high-volume users and provides unlimited notes, attachments, storage, and synchronized devices.

The concern is instead one of long-term ownership and optionality.

Over more than a decade, Evernote has become the container for a substantial portion of my personal digital memory. Moving away from it now is already a significant project. If I simply replace Evernote with another application-centric repository, I may recreate the same problem and face an even larger migration ten years from now.

My objective is therefore not merely to find “another Evernote.” It is to establish a system in which the underlying information remains durable, accessible, searchable, and usable independently of any one software vendor.

Cost contributes to this decision as well. A rough reconstruction suggests that I have probably spent on the order of $1,000–$1,300 on Evernote subscriptions over the past decade-plus. More importantly, Evernote’s current Advanced plan costs $249.99 per year in the United States. Given the size of my account, the lower Starter plan (with a 1,000-note, 1,000-attachment, and 5 GB limit) is not a realistic option. Further, I’ll be maintaining this PKM longer than the seven-years I have until retirement when the costs become my own :)

At today’s price, another ten years of Evernote Advanced would therefore cost approximately $2,500, before future price increases. Retired-Eric will not be paying that much

The Two Leading Alternatives

Obsidian: Files First

Obsidian’s defining advantage is architectural. An Obsidian “vault” is fundamentally a directory of ordinary files on my computer. Notes are stored as non-proprietary Markdown text files, and PDFs, images, and other attachments remain files. Obsidian explicitly describes this model as keeping data locally under the user’s control.

That directly addresses the issue that prompted this evaluation.

If Obsidian disappeared in the future, the underlying collection would remain usable. Markdown can be read by countless applications—or with nothing more sophisticated than a text editor. PDFs remain PDFs. Images remain images. The archive could be indexed by future search products, scripts, databases, or AI systems without first requiring an export from Obsidian.

Obsidian also provides an official Evernote importer. It consumes Evernote’s ENEX exports and converts notes into durable Markdown, with options for controlling metadata and folder structure during migration.

This open-file structure is particularly compelling in the era of AI. The permanent asset would not be “my Obsidian database.” It would be my corpus of files. Search indexes, vector embeddings, AI models, and applications could all be treated as replaceable layers above that corpus.

The disadvantage is synchronization cost at my scale. Obsidian itself is free for personal use, but its official Sync service has storage tiers. Sync Plus provides 10 GB for $8 per month when billed annually and can be upgraded to 100 GB for $16 per month billed annually. My existing 42 GB Evernote collection therefore requires the 100 GB tier if I want the entire archive synchronized through the official service.

That puts the current ten-year cost at approximately $1,920, assuming unchanged pricing.

DEVONthink: Documents First

DEVONthink approaches the problem from the opposite direction. It is a mature Mac-oriented information and document manager designed to handle heterogeneous collections of notes, PDFs, scans, images, records, and other documents.

For a 42 GB archive with thousands of attachments, this matters. My collection is not merely a collection of interconnected text notes; it is also a substantial document repository. DEVONthink’s document-management, OCR, classification, search, and organizational capabilities are particularly well aligned with that workload.

DEVONthink 4 has also become considerably more interesting because of AI. In 2026, DEVONtechnologies added a built-in Model Context Protocol (MCP) server. Compatible AI tools can search DEVONthink databases, read and summarize records, create and update documents, move items, add tags, and perform other operations. Access can be restricted at the database or individual-record level, and DEVONthink provides controls for redacting sensitive content before information is sent to an external AI provider.

That maps directly onto the future use case I care about: asking an AI a question and allowing it to consult my personal archive as part of answering.

DEVONthink also presents a more nuanced lock-in situation than Evernote. By default it imports copies of files into its database, but it can alternatively index files that remain in ordinary Finder folders, maintaining references to those external files while adding DEVONthink’s search and classification capabilities. This creates the possibility of preserving a filesystem-based corpus while using DEVONthink as an intelligence layer over it.

Its cost structure is also favorable. DEVONthink 4 Pro currently costs $199, including two Mac seats, five DEVONthink To Go mobile seats, and one year of updates. Importantly, the purchased Mac application does not stop working when the update period expires. Current customers can extend updates before expiration for $89 for one year, $159 for two years, or $209 for three years.

Using the three-year extension option, ten years of continuously current DEVONthink Pro would cost approximately $826 at today’s prices: $199 initially plus three $209 extensions.

Current Comparison

* At current published prices, excluding general storage, backup, taxes, AI subscriptions/API charges, and future price changes.

AI Changes the Decision

Five years ago I would have evaluated this largely as a synchronization, search, editing, and file-portability problem.

Generative AI adds a new requirement: the archive should be machine-accessible as a coherent knowledge corpus.

The goal is not to upload 42 GB into a chatbot; a better architecture is retrieval-based. (And, as much as possible, retain local control of all files — keeping them out of Skynet’s hands…) Software indexes the corpus, identifies the material relevant to a question, and supplies only those sources to an AI model. Ideally an answer is accompanied by citations back to the original notes or documents.

For example, rather than searching manually through old notebooks, I should eventually be able to ask:

“What decisions have I made over the years about replacing the heating system, and what contractors, estimates, rebates, and technical considerations were involved?”

A useful system should locate information even if I variously described the subject as “furnace,” “HVAC,” “heat pump,” or a manufacturer’s name.

This capability makes open, durable source data more valuable, not less. AI models and retrieval technology will change dramatically over the coming decade. I do not want today’s AI integration to become tomorrow’s new form of lock-in.

It is worth noting that all three platforms are now moving toward AI, including Evernote itself. Consequently, “has AI features” is not a useful differentiator. The better question is how easily the underlying corpus can be connected to whatever the best AI technology is five or ten years from now.

On that criterion, Obsidian’s file-based architecture has a fundamental advantage.

One Repository, Not Two

One possible design would be to divide the collection: Markdown and active notes in Obsidian, large PDFs and archival documents in DEVONthink.

I have rejected that as the preferred architecture.

I do not want personal knowledge management to require remembering that “the recipe is in DEVONthink” while “the blueprint is in Obsidian.” A fundamental requirement is that there be one conceptual home for the entire archive.

It may eventually be reasonable for another application to index or provide an alternate interface to that repository. That is different from splitting ownership of the information across applications.

The choice should therefore be between a unified Obsidian corpus and a unified DEVONthink corpus — or possibly a filesystem corpus for which one of these applications serves as the primary interface.

Preliminary Conclusion

At this stage, Obsidian has the strategic advantage and DEVONthink has the operational advantage.

Obsidian most directly solves the problem that started this evaluation. Its working data consists of ordinary files, providing maximum independence from both Obsidian itself and from today’s generation of AI products.

DEVONthink is arguably the better application for the collection I actually have today: 42 GB of notes, PDFs, images, scans, and attachments. Its lower projected software cost, sophisticated document-management capabilities, and built-in MCP interface make it a very serious alternative.

I do not think the decision should be made from specifications alone.

The next step should be a representative migration pilot, using the same subset of Evernote material in both systems. The test corpus should deliberately contain ordinary text notes, complex formatted notes, PDFs, image-heavy notes, scans, large attachments, tags, internal links, and older material.

The pilot should answer practical questions: How faithfully does each product preserve the archive? How does each perform on the Mac with a substantial corpus? What is the experience on an iPhone and iPad? Are PDFs and images easy to retrieve? How reliable is synchronization? Can an AI answer questions across the test archive and identify the sources it used? And, perhaps most importantly, how confident am I that I could leave the new platform ten years from now without facing another major extraction project?

The objective is not simply to select better note-taking software.

It is to convert more than a decade of accumulated information into a durable, application-independent, AI-ready personal knowledge archive that I can continue using for decades.