LogSeq/Obsidian Vault

How I Think About Knowledge Management

I keep a zettelkasten-style vault in LogSeq. The basic idea: every note is a single atomic idea, written in my own words, connected to other ideas via bidirectional links. Over time the vault becomes a web of interconnected knowledge that gets more valuable the more you add to it, because new notes link to existing ones and create connections you wouldn't have made otherwise.

I migrated everything over from Roam Research into LogSeq. The tool doesn't matter that much -- Obsidian works just as well, and the method is the same either way. Both are local-first markdown editors with bidirectional linking. I went with LogSeq.

There are two types of notes:

• Reference notes -- one per source (book, article, paper, podcast). Summarizes the source, links out to individual insights, holds the metadata.
• Literature notes -- one per idea. A standalone concept written in my own words, 2-3 paragraphs. Cross-linked to related notes and back to the source reference note.

I've built mine up to 350+ literature notes across dozens of books and papers. The real value isn't any individual note -- it's the connections between them.

How OpenClaw Fits In

This is where it gets interesting. My agent has direct read/write access to the vault (via Syncthing if you're running on a separate machine). That means it can both read my existing notes for context and create new ones.

Here's how I actually use it:

Research workflow -- When I find a book, article, or paper worth processing, I send it to my agent. It reads the source, extracts 3-8 key insights, and creates a reference note plus individual literature notes for each insight. It cross-links everything to existing vault notes and logs what it created in the daily journal. A single paper turns into a handful of atomic notes that connect to the rest of my knowledge base.

Writing workflow -- When I'm working on an article or a thread, I'll ask my agent to search the vault for relevant notes on a topic. It pulls together related literature notes and reference notes, which gives me a research foundation to write from. The tweet pipeline also pulls from vault content.

Daily journals -- The Life OS cron jobs write directly to LogSeq journal files. Morning writing logs, daily check-ins, weekly reviews -- they all live in the vault's journal folder and show up in LogSeq automatically.

Vault maintenance -- Over time, notes accumulate and some get under-linked. My agent periodically runs cross-linking sweeps to find connections between notes that should be linked but aren't. See Vault Maintenance.

Setup Prompt

I use [LogSeq/Obsidian] as my knowledge vault at [path].

Set up a zettelkasten system:

- Reference notes: One per source (book, article, paper). Contains:
  summary, key concepts (linked to literature notes), key quotes,
  metadata (authors, date, URL).

- Literature notes: One per idea/insight. Standalone atomic ideas
  written in my own words (2-3 paragraphs). Linked back to the source
  and cross-linked to related notes.

- Daily journals: At [journal path]. Log what was added each day.

When I send you a source to process, create the reference note,
extract 3-8 literature notes, cross-link everything, and update
the daily journal.

Format rules:
- Literature note titles should be full sentences expressing the
  core idea (e.g., "Human training data gives AI a faster start
  but a lower ceiling")
- All cross-linking should be bidirectional where relevant
- No subdirectories -- all pages in a flat folder

Processing Sources

When I send my agent something to process -- a paper, a set of Readwise highlights, an article -- it runs a two-pass pipeline:

Pass 1 (Outline): Read the source and outline 3-8 key insights worth capturing as literature notes. Each one is a full sentence expressing the core idea.

Pass 2 (Write): For each insight, write a standalone literature note (2-3 paragraphs, in my own words, with cross-links). Create the reference note linking to all of them.

The two-pass approach matters. A single pass tends to produce surface-level notes. Outlining first and then writing forces deeper thinking about what's actually worth capturing.

I use sub-agents for Pass 2 so the notes get written in parallel. Opus handles orchestration, Sonnet writes the individual notes. This keeps costs down -- processing a whole book's worth of highlights costs a few dollars instead of ten.

Syncthing

If your agent runs on a different machine than your vault, you need Syncthing to keep them in sync. My vault lives on my laptop where I use LogSeq, and Syncthing keeps it mirrored on the Mac Mini where my agent runs. Notes sync in both directions within seconds.

If everything runs on the same machine, you don't need this -- just point your agent at the vault folder directly.