ObsidiarIA

Obsidian + diary + AI: a concrete, operational operating system for your knowledge work — capture once, the agents file — not a generic "chat with your notes" wiki.

_{Unofficial, community-built template — not affiliated with or endorsed by Obsidian (Dynalist Inc.).}

This is a ready-to-clone Obsidian vault template wired to AI coding agents (OpenCode, Claude Code, Kilo). It turns a pile of notes into a system that tracks project progress, drafts deliverables, runs critical brainstorming, processes your inbox, and never lets a deadline or follow-up slip — because an agent keeps the structure clean for you.

The thesis: most "LLM + notes" setups stop at retrieval ("ask your notes a question"). This one is operational. It is built around how a busy knowledge worker actually works — many parallel threads of work, a heavy meeting load, outputs to produce, follow-ups that must not slip — and it gives the AI a precise contract for where every piece of information lives. It fits a wide range of roles — developers and tech leads, IT and delivery managers, consultants, back-office and HR, researchers — anyone juggling many parallel commitments. For you, day-to-day use is deliberately dull: drop everything into a single inbox and let the agent do the filing.

It ships pre-filled with one illustrative example profile — Daniel Lacemore, a consultant — so you see a working system from minute one; it's only an example, not the intended audience. A one-command wizard (/setup) rewrites that example into your profile, whatever your role.

Note: Daniel Lacemore is a fictional name (an anagram-style placeholder), used only as the template's example profile — it is not a real person.

How it works — capture once, the agent files

The system optimizes for the one habit you can actually sustain: capture everything in the Inbox, and let the agent do the organizing. No tags to pick, no folder to choose, no deciding upfront which project a stray thought belongs to — that filing work is the agent's job, not yours.

1. Capture. A meeting note, a forwarded email, a half-formed idea → a new note in 01_Inbox/. That single folder is the only entry point, and capturing is the only discipline the system asks of you.

Tip — make capture one keystroke. In Obsidian, open Settings → Files and links → Default location for new notes, choose In the folder specified below, and set the folder to 01_Inbox. Now Cmd/Ctrl + N always creates the new note straight in the Inbox — capture without ever thinking about where it goes.

2. Route. When you run /obs.inbox (or /obs.eod), the agent reads each capture, finds or creates the right Activity, and sends every piece of it to its one correct home:

All of it obeys one rule — the anti-scatter principle: a piece of information lives in exactly one place.

3. Read. To get back up to speed on anything, open its Activity (the Master note): current state and pointers at the top, the detail one link away. You never go hunting across folders.

Want a scratchpad the agent won't touch? Tag a note ongoing in the Inbox and it stays put — persistent working notes are never auto-filed.

Why it's different

• Activity-centric, not note-centric. Work is organized around Activities (multi-day units of work). Each Activity is a Master note: a hub of state + pointers. Detail lives in linked child notes. One rule governs everything: a piece of information lives in exactly one place (the anti-scatter routing matrix).
• Tool-agnostic by design. The AI reads and writes plain markdown files. No lock-in to one assistant. The same vault works with OpenCode, Claude Code and Kilo, and stays fully compatible with native Obsidian features (Bases, Dataview, Tasks).
• A small team of specialized agents, not one do-everything bot (see below).
• Single source of truth for AI config. You edit one place (.opencode/) and a sync script regenerates the per-tool configs.
• Light on purpose. No taxonomy sprawl. The agents are told to keep the system simple.

The agents

One primary orchestrator delegates to four specialists, plus a directly selectable brainstorming partner. The codenames are deliberate and memorable; here's what each one does:

The full prompts live in .opencode/.prompts/. Edit there, then re-sync (below).

Folder structure

00_System/      Operating model, dashboard, templates, bases, views, AI config
01_Inbox/       Single raw-capture folder — everything enters here first
02_Daily/       One daily note per day (YYYY-MM-DD)
03_Projects/    One note per project (WBS-coded)
04_Activities/  Activities (Master notes); tasks live as #task checkboxes inside
05_Meetings/    Concise, extraction-friendly meeting notes
06_Notes/       Support notes (type: note) and deliverables (type: deliverable)
07_Accounts/    Client details
08_Attachments/ Files (incl. non-markdown deliverables)
09_People/      People profiles
90_Archive/     Inactive material

The complete property schema and conventions are in 00_System/System Guide.md — the contract both you and the agents follow. The high-level rules the agents load on every session are in AGENTS.md.

Daily workflow

In practice the loop is small and repeatable:

• Morning — /obs.morning. The agent reads yesterday's note, scans your ongoing Activities, surfaces open tasks and blockers, and proposes a Top-3 + focus for the day (it asks you to confirm).
• Through the day — just capture. Meeting notes, ideas, follow-ups → Cmd/Ctrl + N → Inbox, and keep moving. Don't stop to file. Need a structured minute on the spot? /obs.meeting. Want to think a decision through first? Talk to Socrates directly (read-only, dialogic brainstorming).
• End of day — /obs.eod. The agent empties the Inbox into the right Activities, checks off finished tasks, closes resolved open points, updates deliverables and projects, and writes the day's summary. The vault is clean again — and you filed nothing.
• Weekly — /obs.weekly. A retrospective over the week: stagnant open points, overdue tasks, project health, and the priorities for the week ahead.

When a real client output is due, Archie drafts the deliverable and Friday turns it into a slide deck; Melio keeps the vault's structure tidy over time.

The full command set:

Quick start

1. Clone this repo and open the folder as a vault in Obsidian.
2. Install the community plugins it relies on: Dataview, Tasks, Templater, and enable core Bases. (Excalidraw optional, for diagrams.) Point Templater's template folder at 00_System/Templates.
   • Make capture one keystroke: in Settings → Files and links, set Default location for new notes to In the folder specified below → 01_Inbox. Now Cmd/Ctrl + N drops every new note straight into the Inbox.
3. Open the folder with your AI agent (OpenCode recommended; Claude Code or Kilo also work).
4. Run the wizard:

   /setup
   

   It asks for your name, role, language, domain and a few labels, then rewrites the vault to be yours. (Prefer the terminal? See setup/README.md.)
5. Restart your AI tool so the updated agents and commands load.
6. Start your day with /obs.morning or /obs.inbox if you just want to process the inbox notes.

Node.js is required only for the two maintenance scripts (setup/apply-profile.mjs and 00_System/AI/sync-agents.mjs). The vault itself is plain markdown.

Customization

Everything user-specific is captured in setup/profile.yml (name, role, organization type, working language, domain one-liner, project-code label, vault path). The /setup wizard — or node setup/apply-profile.mjs — propagates those values across AGENTS.md, the System Guide, the agent prompts, the commands and the templates. Reset to the shipped example any time with --defaults. Details in setup/README.md.

To change agent behavior, edit the source under .opencode/ and regenerate the per-tool configs:

node 00_System/AI/sync-agents.mjs   # rebuilds .kilo/, .claude/ and CLAUDE.md

The generated files (.kilo/, .claude/, CLAUDE.md) carry an AUTO-GENERATED header — don't edit them by hand.

Plugins & integrations

• Required Obsidian plugins: Dataview, Tasks, Templater, Bases (core). Optional: Excalidraw.
• MCP servers (search, vendor docs, diagrams) are not versioned here: configure them at your tool's user scope. They often hold API keys and machine-specific paths. The agents degrade gracefully when an MCP is missing. See 00_System/System Guide.md §7 for the matrix and the secret-hygiene rule (use env-var interpolation, never cleartext keys).
• Skills: see SKILLS.md for the bundled and the optional external skills (with attribution).

License & credits

• Code and configuration: MIT — see LICENSE.
• The optional external skills referenced by the agents belong to their respective authors — see SKILLS.md.
• Agent codenames are a playful nod to fictional assistants; the system itself is your own.