Knowledge Forge (WIP)

An Obsidian vault for research + startup work. It is not a note manager; it is a pipeline:

Capture → Source → Concept → Idea → Experiment / Project → Writing / Output

Raw material (papers, books, videos, web pages, GitHub repos, movies, market research) gets forged into your own knowledge and output.

Folder structure

00_Inbox/          Capture here, do not classify
01_Projects/       What I am pushing forward (project dashboards, not encyclopedias)
02_Sources/        What other people said (raw material → my source note)
  Papers/          Highlights/ (ZotLit-managed) + Notes/ (my paper notes)
  Books/           Highlights/ (KOReader-managed) + Notes/
  Articles/        Web article notes (raw clip lives in 09_Resources/Clippings)
  Videos/          Video notes (Media Extended)
  Repositories/    GitHub repo notes
  Movies/          Movie notes (TMDB)
  Podcasts/  Reports/
03_Concepts/       What I already know (one concept per note)
04_Entities/       People / Company / Products / Things (Entity, People templates)
05_Experiments/    What I actually tried (EXP001 - ...)
06_Ideas/          New hypotheses
07_Writing/        Papers / reports / articles; Blog/; Blog/
08_Meetings/       Supervisor / cofounder / customer
09_Resources/      Clippings/ (Web Clipper output), Tools/, How-to/, TroubleShoot/, Collection/
10_Business/       Reusable business knowledge: Market / Strategy (see its README)
11_Daily/          Daily note = time log, archived as YYYY/MM
12_Admin/          Admin / paperwork
13_Finance/        Accounting / Investing / Researching
90_External/       Symlinks to external git repos; readable by Obsidian, git-ignored
99_Archive/        Finished projects, rejected directions
Attachments/       Diagrams/{Mermaid,Excalidraw}  Image/  Paper/  Export/
Scripts/           QuickAdd user scripts
Templates/         Note templates + Mermaid/ diagram snippets + Writing/ paper sections
.obsidian/         Core + plugin settings (data.json only); plugins/knowledge-forge-ai = optional submodule

Where does a note go?

The question it answers	Folder
What did I just capture?	00_Inbox
What am I pushing forward?	01_Projects
What did someone else (paper/book/video/page) say?	02_Sources/*
What do I know about this field?	03_Concepts
What did I try?	05_Experiments
What new hypothesis do I have?	06_Ideas
What am I producing?	07_Writing
This company / this market / long-term strategy?	04_Entities/Company, 10_Business/{Market,Strategy}
What happened today?	11_Daily

Rule: the thing you are doing goes in 01_Projects; the reusable knowledge it produces settles into 02–10. The example notes in each folder show this with one sample project, LLM Reasoning Study.

Root files

• Dashboard.md — today / overdue / next 7 days / active projects / reading queue
• Agenda.md — org-agenda style Tasks queries
• Kanban.md — query-based board; tasks stay in their original notes
• style-settings.json — Style Settings plugin config
• Templates/README.md — every note type, placeholder syntax, Mermaid pack
• 10_Business/README.md — business folder guide
• .gitignore — excludes 90_External/

Plugins

Minimal set. Tasks queries tasks; Dataview queries note metadata. Keep them separate.

Role	Plugin	Required
org-capture / custom commands	QuickAdd	yes
TODO / scheduled / deadline / agenda / kanban	Tasks	yes
Metadata queries (papers, projects, experiments)	Dataview	yes
Zotero papers + annotations	ZotLit (+ ZotLit Companion in Zotero)	for papers
KOReader book highlights	KOReader Highlights Importer	for books
Video notes / timestamps	Media Extended	for videos
Web pages	Obsidian Web Clipper (browser extension)	for articles
Diagrams	Mermaid Next, optional Excalidraw	optional
Pomodoro	any minimal timer (PomoBar)	optional
AI edit / note creation, local-first (experimental)	Knowledge Forge AI (git submodule)	optional

Core plugins: Backlinks, Daily Notes, Templates, Properties, Quick Switcher, Bases.

.obsidian/ ships the core and plugin settings (data.json) but not the plugin code: install each plugin from Community plugins and the settings are picked up.

Knowledge Forge AI (optional, experimental) lives at .obsidian/plugins/knowledge-forge-ai as a git submodule, not fetched by a plain clone:

git submodule update --init .obsidian/plugins/knowledge-forge-ai
cd .obsidian/plugins/knowledge-forge-ai && npm ci && npm run build

Then enable it in Community plugins. Defaults to a local Ollama endpoint (qwen3:8b).

Plugin settings

Templates — Settings → Templates → Template folder: Templates.

Daily notes — Settings → Daily notes:

• New file location: 11_Daily/{{date:YYYY}}/{{date:MM}}
• Template: Templates/Daily Note

Tasks — defaults are fine. Use 📅 for deadline and ⏳ for scheduled: - [ ] Read Ansor 📅 2026-09-20 ⏳ 2026-09-18. Enable "Auto-suggest" for the emoji picker.

Dataview — enable "Enable JavaScript queries" is not needed. Queries rely on frontmatter keys: type (project / source / idea / experiment / daily), status, source_type, projects, updated.

QuickAdd — see Build the flow below.

ZotLit — Settings → ZotLit:

• Literature note folder: 02_Sources/Papers/Highlights

• Templates → Note filename: create editable template file, use 02_Sources/Papers/zotlit-filename.liquid.md ({{ zt.title | truncate: 100 }} - {{ zt.citationKey }}{% suffix %})

• Templates → Note content: use 02_Sources/Papers/zotlit-content.liquid.md

• Templates → Frontmatter: add managed fields so Create Paper Note can inherit them:

  Key	Expression
  title	zt.title
  authors	zt.creators | map: "fullName"
  year	zt.date | date: "%Y"
  citekey	zt.citationKey
  doi	zt.DOI
  venue	zt.publicationTitle
  item_type	zt.itemType

  Use ZotLit's Template Data Explorer to check which zt.* fields your Zotero items actually have.

• Zotero side: install ZotLit Companion; Better BibTeX for stable citation keys.

KOReader Highlights Importer — Settings → KOReader Highlights Importer:

• Scan path / data folder: KOReader's docsettings directory, e.g. ~/.config/koreader/docsettings
• Highlights folder: 02_Sources/Books/Highlights
• KOReader side: Settings → Document → Book metadata location → ./docsettings (keeps .sdr out of the Calibre library; use Move book metadata to migrate existing .sdr).

Media Extended — Settings → Media Extended:

• Note folder / default note location: 02_Sources/Videos
• Timestamp / screenshot insert into the active note. Create Video Note writes mx-uid so "Open note" from the player jumps back to the right note.

Obsidian Web Clipper (browser) — in your Article template:

• Note location: 09_Resources/Clippings
• Note name: {{title}}
• Keep url, author, published, site properties; Create Article Note reads them.

Excalidraw (if used) — drawing folder: Attachments/Diagrams/Excalidraw.

CSS — put tasks.css in .obsidian/snippets/, enable under Settings → Appearance → CSS snippets.

Build the flow

Step 0: folders and templates

1. Create the folder tree above.
2. Copy Templates/ and Scripts/ into the vault root.
3. Apply the plugin settings above.

Step 1: Capture (org-capture)

QuickAdd → Add Choice Capture Inbox, type Capture, file 00_Inbox/Inbox.md, append to end. Bind Ctrl+Alt+C.

Rule: never classify while capturing. Concept / Idea is decided later.

Step 2: register the QuickAdd scripts

One Choice per script: type Macro → Macro Builder → Add → User Script → pick Scripts/xxx.js → enable Add to command palette.

Choice	Script	What it does	Needs
Create Project	create_project.js	Prompts name / area / repo path → 01_Projects/X.md + 90_External/X symlink (repo-less variant available)	—
Create Paper Note	create_paper_note.js	Pick a file in Papers/Highlights → renders Templates/Paper.md to Papers/Notes/{author} {year} - {title}.md, inherits metadata, links back	ZotLit
Create Book Note	create_book_note.js	Pick a file in Books/Highlights → Books/Notes/	KOReader Importer
Create Video Note	create_video_note.js	YouTube / Bilibili URL or local path → 02_Sources/Videos/	Media Extended
Create Article Note	create_article_note.js	Pick an existing clip, or start from a URL → 02_Sources/Articles/	Web Clipper
Create GitHub Repo	create_github_repo.js	owner/repo or URL → GitHub API metadata → 02_Sources/Repositories/, optional symlink of local clone into 90_External	optional GitHub token (script settings)
Create Movie	create_movie.js	TMDB search / IMDb id → 02_Sources/Movies/	TMDB Read Access Token (script settings)
Create Diagram	create_diagram.js	Mermaid: pick template, write standalone file in Attachments/Diagrams/Mermaid/ or embed inline; Excalidraw: delegates to its command	Mermaid Next / Excalidraw
Insert Mermaid	insert_mermaid.js	Insert a snippet from Templates/Mermaid/* at the cursor	Mermaid Next

Scripts with a settings block (paper, github, movie) expose their folders/tokens in QuickAdd → the Choice's gear icon.

Template Choices (type Template): Create Idea → Templates/Idea.md into 06_Ideas; Create Experiment → Templates/Experiment.md into 05_Experiments; likewise Concept, Meeting, Company Research (→ 04_Entities/Company), Market Research.

Step 3: hotkeys

High-frequency actions get a key; everything else goes through the QuickAdd menu:

Action	Linux / Windows	macOS
Capture Inbox	Ctrl+Alt+C	Cmd+Opt+C
Create Project	Ctrl+Alt+P	Cmd+Opt+P
Create Idea	Ctrl+Alt+I	Cmd+Opt+I
Create Experiment	Ctrl+Alt+E	Cmd+Opt+E
Insert Mermaid	Ctrl+Alt+M	Cmd+Opt+M
QuickAdd Menu	Ctrl+Alt+Space	Cmd+Opt+Space

Step 4: source pipelines

Papers: Zotero → ZotLit → Obsidian

Browser / arXiv / Scholar
  ↓  save to Zotero (PDF + metadata + Better BibTeX key)
Zotero
  ↓  read & highlight the PDF in Zotero
ZotLit: "Open literature note quick switcher" → pick the item
  ↓  creates/opens 02_Sources/Papers/Highlights/<Title - citekey>.md
      (managed region: frontmatter, annotations, Zotero back-link;
       re-run "Update literature note metadata" after new highlights)
QuickAdd: Create Paper Note → pick that Highlights file
  ↓  02_Sources/Papers/Notes/<Author Year - Title>.md  (your own reading note)
Concepts / Ideas / Experiments → Project

• PDFs never enter the vault; the zotero: link in frontmatter opens the item.
• Do not write your thoughts in the Highlights file — ZotLit owns it. Write in the Notes file.

Books: Calibre → KOReader → Obsidian

Calibre   manages EPUB / PDF / metadata
  ↓  open the book with KOReader (Calibre: set KOReader as external viewer)
KOReader  read, highlight, add notes → written to .sdr in ~/.config/koreader/docsettings
  ↓
Obsidian: "Scan KOReader Highlights" (check what it found)
Obsidian: "Import KOReader Highlights"
  ↓  02_Sources/Books/Highlights/<Book>.md  (managed; re-import merges)
QuickAdd: Create Book Note → pick that Highlights file
  ↓  02_Sources/Books/Notes/<Book>.md  (your own reading note)
Concepts / Ideas

Web pages

Web Clipper → 09_Resources/Clippings/<title>.md → Create Article Note → 02_Sources/Articles/

or run Create Article Note directly from a URL (it clips first, then creates the note).

Videos: open the URL / file in Media Extended, run Create Video Note, then insert timestamps and screenshots into that note while watching.

GitHub repos, movies: the scripts fetch metadata and fill the template.

Diagrams: full Mermaid snippet set in Templates/Mermaid/; Create Diagram handles Mermaid and Excalidraw, output in Attachments/Diagrams/.

Step 5: external projects

For code that lives in a GitHub repo but whose docs you want to edit in Obsidian:

1. Create Project with the repo path → it runs ln -s <repo> 90_External/<name>.
2. .gitignore already excludes 90_External/, so the vault repo never commits the linked project.
3. Link from 01_Projects/<name>.md to 90_External/<name>/docs/....

Step 6: tasks and agenda

• Tasks live in their notes (project / daily / source). Agenda.md, Dashboard.md, Kanban.md are pure queries.
• Dataview views depend on frontmatter (type, status, updated); keep the templates' frontmatter intact.

Daily loop

Morning  open project note → Next Actions → pick 1–3 things → Agenda.md
Reading  capture to Inbox first → worthwhile items go through a source pipeline
         → distill Concepts → Ideas emerge
Evening  Daily note Work Log, one section per project, linked to Concepts / Ideas
Weekly   empty 00_Inbox; update project status / updated; move finished work to 99_Archive

How much to write per paper: A-tier (core) — full note; B-tier (related) — core idea + relation to my work; C-tier (skimmed) — one line.

Target knowledge graph

                    [[Research Question]]
                           ↑
[[Paper A]] → [[Concept]] ← [[Paper B]]
                  ↓
              [[Idea]]
                   ↓
              [[Experiment]]
                   ↓
                [[Paper]]

Three questions the vault must answer: what did others find → Sources; what do I know → Concepts; what will I contribute → Ideas + Experiments + Writing.