Summary of "How I Use Obsidian + Claude Cowork to Run My Life"
Summary of Technological Concepts / System Setup
The video explains a personal “AI operating system” that combines:
- Obsidian: a place to store the user’s thinking as Markdown notes
- Claude Co-work: an AI tool that can access those notes through controlled folder permissions
The goal is future-proofing—so the user can swap AI tools later (e.g., Claude → Codex/open models) without being locked in. This is achieved by keeping the key “translation layer” inside the user’s own notes.
Core architecture: 3-layer “AI OS”
-
Ideaverse (inner layer) The user’s own thoughts and knowledge stored in Obsidian.
-
Translation layer (middle) “Maps and manuals” inside Obsidian that instruct the AI how to:
- navigate and understand identity
- follow the user’s skill instructions safely and consistently
-
External AI tools (outer layer) In the example, Claude Co-work, which reads/modifies the folder the user grants access to.
Obsidian as a Portable Knowledge Vault
- Obsidian is described as a folder (“vault”) of Markdown text files.
- Because Markdown is standard, notes can be opened by other apps and tools, meaning the system is not dependent on Obsidian remaining available.
- The vault is organized into primary folders (using ACE):
- Atlas: knowledge, reference, ideas
- Calendar: time-based notes (daily notes, journals, meetings)
- Efforts: projects, tasks, productivity work
Note-linking and graph view
- Uses Obsidian’s double-bracket linking to connect notes.
- This builds relationships that show up in graph view, reinforcing associations over time for:
- better recall
- better internal linking
Claude Co-work Integration (Outer Layer)
Permissions / file access model
- The user installs Claude desktop and uses Co-work.
- In Co-work, they select a local folder and allow access.
- Claude can then:
- read files
- modify
- move/rename
- create new files
Key idea: Claude only needs folder access, not repeated uploads or re-explaining context each session.
Model routing strategy inside Co-work
- Sonnet: default for most tasks
- Opus: used for “meaty” work—complex synthesis and deep research (higher quality priority)
- Haiku: used rarely; mentioned mainly for Obsidian web clipper summarization / lighter tasks
The system is designed for repeated use with the right model per job type.
Why Claude specifically (as described)
- Claude claims it doesn’t train on the user’s data (described as “non-negotiable”).
- Data retention is described as a rolling 30-day window.
- Claude is described as less restrictive than tools that attempt to entrench users into proprietary ecosystems.
The “Translation Layer” (Middle Layer) Inside Obsidian: The AIOS Folder
A critical warning is included: if the AI is pointed directly at a very large vault (example given: ~17,000 notes), it may sample/cherry-pick and produce inaccurate results because it can’t reliably navigate.
Fix: create an AIOS folder (separate from the Ideaverse) inside the main vault.
Key translation files (“maps and manuals”)
-
me.md(portable identity)- A Claude-agnostic identity prompt describing:
- who the user is
- how the user wants the AI to work with them
- Placed where Co-work points (in a knowledge folder root).
- Designed to avoid lock-in to Claude-specific files (contrasted with Claude-specific files like
Claude.md/ “Claude MD”). - Used during session startup so rules are consistently loaded.
- A Claude-agnostic identity prompt describing:
-
Vault map (navigation / table of contents)
- A structured guide for how the AI should move through notes.
- Enables the AI to pick relevant files/collections without scanning everything.
- Improves precision by loading only what matters.
-
Skill map (when/how to use skills)
- Documents what “skills” exist and when the AI should use them.
- Strong separation principle:
- skills are stored in the user’s own Obsidian notes
- not stored “inside Claude” or any vendor system
- The AI can create/review skills, but they remain owned by the user.
Session Startup Prompt + Hotkey Workflow
The user recommends firing an initial prompt at the start of every new session to:
- read
me.md - review the vault map and skill map
- confirm context is loaded and wait for instruction
They also suggest using a hotkey / text replacement tool (e.g., TextExpander) to paste the startup instruction quickly, reducing “AI amnesia” between sessions.
Examples of Systemized Skills and Workflows
The video emphasizes scheduled and systemized AI “skills,” organized into repeatable systems. Examples include:
Daily Trident system
- Daily brief (morning, scheduled 6:00 a.m.)
- Includes:
- weather
- what happened yesterday / progress
- what’s open
- important emails/logistics
- travel and deadlines
- momentum on major projects (e.g., a book)
- “what you might be forgetting”
- Contains open-note sections where the user can add notes during the day (e.g., phone → later on computer)
- Includes:
Sources mentioned:
- Gmail
- Calendar
- Teams
- ClickUp (tasks/projects/communications)
- Ideaverse notes
- Daily log (scheduled throughout the day)
- An “interstitial journal” for check-ins like what was accomplished
- Feeds forward into the next day’s daily brief
Learning / review / feedback systems
- Sherpa system: helps map and accelerate learning a new topic.
- Weekly review system: adds another lens to ensure nothing slips.
-
Rock tumbler system: fast feedback via open-ended questions using an IDI framework:
- imagine / discern / integrate and Ericcson’s “fast feedback principle of excellence.”
-
Chronicler system: preserves ideas/conversations by saving them verbatim into a history folder; may optionally run a summarizer skill to produce both transcript and summary.
- Janitor system: ongoing maintenance/self-repair of the AIOS (reliability and practicality emphasized).
- Courier system: shares notes with a team while sanitizing personal info and placing content in the right team location; can also include linked-note summaries.
Guidance / Learning Path Provided
- Viewers are pointed to a description link / guide and a QR code that:
- explains how to set up basics and advanced parts
- guides creation of the map files (
me.md, vault map, skill map) - may include an ACE framework
- Message: you don’t need a fully built Ideaverse—you can start with existing notes, using the guide as a bootstrap.
Main Speakers / Sources (As Stated)
- Nick Milo (main speaker; creator of the workflow)
- Claude Co-work (Claude desktop app / Claude models), used as the external AI tool
- Ericcson / K. Anders Ericsson (referenced for the “fast feedback principle of excellence”)
- IDI framework (referenced as the structure used in one skill; attributed by name in subtitles)
Category
Technology
Share this summary
Is the summary off?
If you think the summary is inaccurate, you can reprocess it with the latest model.