Summary of "Claude just killed ALL Note-Taking Apps. Here is proof."

Main thesis

The era of traditional PKM apps (Notion, Obsidian, Heptabase, Roam, etc.) is ending because AI agents running over simple local folders plus a small local database offer more flexible, owned, and customizable Personal Knowledge Assistance (PKA). Keep your data in plain files and a local DB, and let an AI orchestrator + specialist agents manage, index, and present it. You can swap the underlying LLM (Claude, Gemini, ChatGPT, or local LLMs) later.

Core technology and product features demonstrated

Claude Code / Claude Desktop / Claude Co-work used as the head AI (orchestrator) and to run agent teams inside a local folder. Terminal-based usage is shown as an effective workflow.
Folder-based data store (example: “PKA demo”) containing:
- owner’s inbox (AI outputs for review)
- team inbox (files you drop for AI to process)
- team folder / roster (agent definitions and profiles)
- hidden .claude folder (.md files) with agent definitions and memory
Agents (named personas) defined as readable .md files, editable in any text editor:
- Larry — orchestrator / first point of contact
- Pax — senior researcher
- Nolan — HR / hires AI agents
- Sable — front-end / PKM app developer
Plan mode vs. direct edits: Plan mode produces a stepwise implementation plan and asks clarifying questions before acting.
Massive context windows and local control: demonstrations used a large-context Opus model (1M+ window) enabling orchestration across many documents.
SQLite database: the AI creates a local DB (schema for journals, meetings, contacts, file index, tasks) to index and power search over local data.
File ingestion and processing: AI can OCR, detect language, categorize, rename, and folder PDFs/images dropped into the team inbox.
UI/front-end generation:
- Option A: hire an AI developer agent to generate a full local app (runs on localhost).
- Option B (preferred minimal solution): create a simple static HTML viewer that visualizes the SQLite DB.
Remote control: access the same Claude Code session from phone via remote control; terminal + Claude mirrored in the mobile app.
Background tasks and concurrency: run agents in background (Control+B), interrupt with Escape, continue with “continue”; multiple parallel terminals/agents scale work.
Backup & sync: the local folder can be backed to iCloud, Dropbox, or Google Drive. Use local execution or local LLMs for greater privacy.
Cost/context notes: the presenter references Claude Max 20X ($200/mo), Pro ($20/mo), and 5X Max ($100/mo) depending on token usage.

Practical tutorial — step-by-step workflow

Create an empty folder (e.g., “PKA demo”) on your desktop.
Open Terminal, copy the folder path (Option + right-click on Mac), and cd into the folder.
Launch Claude (Claude Code / Desktop) and grant folder access.
Use a natural-language prompt to instruct an orchestrator agent (Larry) to set up guardrails and create team members (Pax, Nolan, etc.).
Let Claude create the hidden .claude folder and agent .md files; review and edit those .md files if needed.
Use Plan Mode to confirm the approach and ask clarifying questions before execution.
Ask the system to create a local SQLite database schema for journals, meetings, contacts, files, interactions.
Drop scanned PDFs / images into the team inbox; ask the AI to OCR, categorize, rename, and index them into the database.
Choose a front-end:
- Option A — ask AI to build a simple static HTML viewer that reads the SQLite DB and opens in any browser (no server).
- Option B — ask AI to build a fuller local app (runs on localhost:3000); revert if overkill.
Use remote control on your phone to check status and message the AI team.
Iterate: adjust agents, guardrails, and folder structures per domain (business vs. personal folders).

Analysis and arguments presented

Lock-in problem: PKM apps often impose developer-defined schemas and plugins, which can lock you into formats or limit export (example grievance with Heptabase).
Ownership & portability: keeping plain files + a local DB gives real ownership and lets you swap the LLM “brain” later.
Individual empowerment: each person benefits by learning to operate their own AI team; productivity gains come from mastering prompts and workflows.
Tool-agnostic productivity: develop mental models and methodologies (e.g., the referenced I core methodology) instead of copying templates that lock you into others’ workflows.
Tradeoffs:
- AI can overcomplicate (e.g., hiring an agent to build a Notion-clone).
- Simpler front-ends are often sufficient.
- Guardrails and iterative review are essential to avoid runaway complexity.
Security and privacy: you can run everything locally or use local LLMs; backup/sync is optional but choose based on privacy needs.

Practical product/feature tips and shortcuts

Show hidden files on macOS: Shift + Command + .
Copy folder path on macOS: Option + right-click → “Copy as Path”
Terminal controls:
- Control + B to run in background
- Escape to interrupt
- Type “continue” to resume
Toggle voice mode in Claude Code: slash (/) → voice, hold space to speak
Workflow tips:
- Ask agents for clarifying questions.
- Require agents to route tasks through the orchestrator (avoid direct assignment unless desired).
- Prefer small, focused folders/projects (separate business and personal) rather than one huge folder.

Guides, tutorials, and references

The video itself is a step-by-step tutorial/demo showing how to build a Personal Knowledge Assistant with Claude Code and a local folder.
Previous channel videos: reviews and history of note-taking apps (Roam, Obsidian, Notion, Mem, Tana, Heptabase).
Prior video explaining why Obsidian isn’t truly local / warns about format lock-in.
Workshop / membership: “building your personal AI team” workshop recordings and live workshops (paid membership) for hands-on setup help.
I core methodology: productivity teaching referenced (tool-agnostic approach).

External services, models, and tools mentioned

Claude (Code, Desktop, Co-work)
Opus model (large-context)
Gemini, ChatGPT
Perplexity API, Gemini Nano
Banana API (image generation)
SQLite, local LLMs
iCloud, Dropbox, Google Drive (sync)
PKM apps for comparison: Notion, Obsidian, Heptabase, Roam Research, Mem, Tana, Craft

Claim / Conclusion

For 2026, a small set of local files + a local SQLite DB plus an AI orchestrator and specialist agents (run via Claude Code or comparable LLM) can replace conventional PKM apps, providing better ownership, flexibility, and scalability. The presenter demonstrates this with a hands-on example including UI generation and file ingestion.

Main speakers / sources

Thomas — video creator / presenter demonstrating the workflow.
Claude — the AI platform used (Claude Code, Claude Desktop, Co-work).
Example agent names used in the demo: Larry (orchestrator), Pax (researcher), Nolan (HR), Sable (developer).
Co-founder referenced: Paco Cantero.
Other referenced services/models: Gemini, ChatGPT, Perplexity.