Skip to content

PKM Community Research

  • Semantic search is the #1 pain point — users hate that default search is “barely substring matching” and want meaning-based retrieval like Google
  • Smart Connections plugin (4,500+ GitHub stars, 800K+ downloads) shows massive demand for AI-powered semantic search that finds related notes automatically
  • Users want proactive surfacing — the tool should show you relevant notes without you having to search
  • As you type, show related notes from your past
  • “Don’t make me remember tags or search manually — retrieve relevance for me”
  • This is the killer feature that new tools like Mem.ai and Constella are building around
  • Collaborative note editing and sharing is on Obsidian’s official roadmap
  • Teams want to effortlessly discover related project information without manual curation
  • Obsidian’s mobile app is praised, but Roam’s mobile “was a flop”
  • Quick capture from anywhere, including voice
  • Navigation and note retrieval on mobile remains frustrating compared to Apple Notes
  • Calendar view for bases/databases is highly requested (on Obsidian roadmap)
  • Kanban views for task management
  • Better daily notes workflows
  • Graph view is popular but often “completely unusable if you have a large number of notes”
  • Users want local graphs that update automatically and help navigate connections
  • Obsidian’s graph view is praised as “incredible” with filters, groups, and customization
  • Roam/Logseq excel here — every block is its own addressable unit
  • Obsidian’s block references feel “hacky” with weird identifiers
  • People want to link and embed atomic note elements across notes easily

“Every attempt at PKM has landed me in the same place: a huge mess”

  • Users collect thousands of notes that never get reviewed or connected
  • Quick capture creates massive unprocessed backlogs
  • “When there are 200+ notes on a topic, distilling useful information becomes impossible”
  • “The more a tool can do, the more it demands from you”
  • “You’re not thinking — you’re maintaining”
  • Obsidian infamous for the “tinker loop”: install plugins, configure CSS, style graphs, never actually take notes
  • “Every feature is another decision”
  • Beautiful dashboards that never get looked at again
  • “High-maintenance digital gardens that require more watering than they return in fruit”
  • Systems that look helpful but waste time
  • Constantly switching between Notion, Obsidian, Roam, Bear, Evernote, Anki
  • Searching for the “perfect solution” and losing momentum each time
  • “Every new note opens up a decision tree: Should this be a page? A subpage? A template?”
  • “The last thing you need mid-idea is five technical questions about structure”
  • “Most PKMs are great at capture… but almost none are built for recall”
  • “I can’t find what I need. Or worse — I forget that I already captured something, because it never surfaces again”
  • File system timestamps get changed when copying/syncing
  • No official API for accessing Obsidian Sync data
  • LogSeq still lacks a sync service

How People Want AI to Integrate with Their Notes

Section titled “How People Want AI to Integrate with Their Notes”
  • Natural language queries: “What did I learn about X?” or “Find notes related to this project”
  • Mem.ai’s Chat feature lets you “interact with your notes using natural language”
  • “Ask questions, edit notes, organize collections, recall details from your knowledge base”
  • AI should understand meaning, not just keywords
  • Smart Connections plugin uses 1,536-dimensional vectors (OpenAI embeddings) to find semantically similar notes
  • Real-time suggestions as you type
  • “Heads Up” features that surface related notes while you work
  • “Future apps will anticipate information needs based on context, schedule, and work patterns”
  • Show relevant past notes without manual search
  • 30% reduction in time organizing notes after meetings with AI summarization
  • Auto-generate meeting recaps, action items
  • Summarize long notes highlighting key points
  • AI Note Tagger plugin is trending in 2025
  • Auto-categorization and tagging
  • Concept mapping and knowledge graph generation
  • Voice-to-text transcription for brain dumps
  • Capture notes verbally, have them processed into structured text
  • AI that requires manual setup/configuration
  • Another system to maintain
  • AI that doesn’t understand personal context (just general training data)

Frontmatter & Metadata Patterns People Use

Section titled “Frontmatter & Metadata Patterns People Use”
  1. Minimal by design: “Only add properties you actively use rather than including unused fields”
  2. Purposeful metadata: “If a property is not used to answer a relevant question, it should not be included”
  3. Consistency matters: Use standardized, lowercase property names to avoid duplication
PropertyPurpose
aliasesEasier linking (e.g., “Chuck” links to “Chuck Mangione”)
typeCategorize notes: journal, people, place, organization, recipe
tagsOrganization and DataView queries
statusTrack note state (fleeting, evergreen, discharged)
created_atWhen note was created (filesystem dates are unreliable across syncs)
projectProject association
authorFor reference notes
urlQuick access to source
  • Nested YAML: Some users create hierarchical metadata (e.g., event.location, event.date) to separate file-level properties from content-level properties
  • Type-based templates: Different note types (book, person, event) have different required properties
  • Genesis tracking: Track whether note is original, derived, or external to measure how much is your own thinking
  • Adding creation_date and modification_date when you don’t use them
  • Including blank/unused properties in notes
  • Over-engineering metadata schemas before you have actual use cases

What Makes Obsidian/Roam/Logseq Successful

Section titled “What Makes Obsidian/Roam/Logseq Successful”
  1. Local-first: Plain Markdown files stored on your computer, not in the cloud
  2. Privacy: No telemetry, no data harvesting — you own your data completely
  3. Data portability: Notes work in any text editor; you’re never locked in
  4. Plugin ecosystem: 2,692 plugins, 97.7M+ total downloads in 2025
  5. Customization: Themes, CSS snippets, configurable everything
  6. Graph view: Best-in-class visualization of note connections
  7. Mobile app: Full-featured iOS app with plugin support
  8. Business model: Free core app, paid add-ons (Sync, Publish) fund development without VC
  9. Community: 110,000+ Discord members, grassroots growth to 1M+ users
  1. Pioneered bidirectional linking in mainstream PKM
  2. Block-level references: Every block is addressable and linkable
  3. Clean UI: Live preview rendering as you type
  4. Daily Notes first: Opens to today’s page, captures everything there
  5. Queries: Powerful embedded queries that surface related blocks
  6. Task management: Built-in todo cycling with keyboard shortcuts
  7. Outliner DNA: Perfect for hierarchical, structured thinking
  1. Best of both worlds: Roam’s outliner UX + Obsidian’s local Markdown files
  2. Free & open source: No subscription required
  3. Block references: Same power as Roam
  4. PDF annotation: Built-in support
  5. Task states: More granular than Roam (Now/Later/Done)
  6. Query table view: Clean display for query results
  7. Privacy: Local storage like Obsidian
FactorWhy It Matters
Bidirectional linkingDiscover relationships organically without upfront organization
Plain text foundationFuture-proof, portable, works anywhere
ExtensibilityCustomize to your workflow
Active communityPlugins, themes, shared knowledge
Philosophy over featuresOpinionated about how knowledge works

  1. Reduce friction at capture time — zero decisions, zero structure required
  2. Make recall automatic — don’t make users search; surface relevant content proactively
  3. AI should understand meaning, not just keywords
  4. Local-first wins trust — privacy and data ownership matter enormously
  5. Avoid feature bloat — every feature is a decision users must make
  6. Block-level granularity matters for serious knowledge work
  7. The graph view is cool but often unusable — local/filtered graphs are what actually help
  8. Mobile is still unsolved for most tools
  9. Don’t make users become janitors of metadata
  10. The real problem is retrieval, not capture — most tools fail at surfacing past knowledge