Intermediate 📅 Last Updated: July 1, 2026 ⏱️ 14 min read 📂 Obsidian + AI

Obsidian AI Workflow: Notes, Projects, Prompts, and Agent Memory

Who This Is For

Read this if: You use Obsidian (or want to), you run local AI models, and your notes are a mess. You want a system that makes your vault queryable by AI — not just a dumping ground.

Skip if: You have never used Obsidian. Start with Obsidian + AI basics first.

What You Need

• Obsidian installed (free)
• Ollama running locally (see our install guide)
• Optional: Open WebUI or Goose for agent workflows
• 10 minutes to set up the folder structure

The Six-Folder Vault Structure

Every folder has a number prefix so Obsidian sorts them in workflow order — capture first, knowledge last.

MyVault/
├── 00-Inbox/              ← Quick capture, unsorted
├── 01-Projects/           ← Active work, one folder per project
│   ├── website-redesign/
│   └── research-paper/
├── 02-Agents/             ← Agent configs and memory files
│   ├── coding-assistant.md
│   ├── research-agent.md
│   └── _memory/           ← Agent memory files (JSON/MD)
├── 03-Prompts/            ← Reusable prompt templates
│   ├── summarize.md
│   ├── code-review.md
│   └── daily-review.md
├── 04-Research/           ← Source material, web clips, PDFs
│   ├── papers/
│   └── web-clips/
├── 05-Knowledge-Base/     ← Permanent, polished notes
│   ├── concepts/
│   ├── decisions/
│   └── how-tos/
├── Templates/             ← Obsidian templates
├── .obsidian/             ← Obsidian config (auto)
└── README.md              ← Your vault map

Why Numbered Folders?

Obsidian sorts alphabetically. Numbers force the workflow order: you see Inbox first (process it), Knowledge Base last (it's done). This removes friction.

How Notes Flow Through the Vault

1. Capture → Note lands in 00-Inbox (quick thought, voice memo, web clip)
2. Process → Daily review moves it to the right folder
3. Act → Project notes live in 01-Projects while active
4. Archive → When a project finishes, extract lessons to 05-Knowledge-Base

Agent Memory: How AI Remembers Your Vault

This is where it gets powerful. Your 02-Agents/ folder stores agent configuration files. Each agent references a memory file in 02-Agents/_memory/ that persists context between sessions.

## 02-Agents/coding-assistant.md
---
name: coding-assistant
model: qwen2.5-coder:14b
memory_file: _memory/coding-context.json
---
You are my coding assistant. You have access to my project
notes in 01-Projects/. Check 05-Knowledge-Base/decisions/
for my coding standards before suggesting changes.

For a deep dive on how agent memory works with Obsidian, see Guide #11: Using Obsidian as Long-Term AI Memory.

Prompt Templates That Work

Store prompts as Markdown so you can copy-paste into any AI tool. Here is a real one from our vault:

## 03-Prompts/summarize.md
---
purpose: Summarize long notes for the knowledge base
tags: [prompt, summarize]
---
Summarize the following note in 3 bullet points, then
extract any action items into a checklist at the bottom.

Note content:
{{content}}

The Daily Review Template

This is the engine of the whole system. Run it once a day to process your Inbox and keep the vault clean.

## Templates/daily-review.md
---
date: {{date}}
---
# Daily Review — {{date}}

## Inbox Processing (00-Inbox)
- [ ] Review all notes in 00-Inbox
- [ ] Move each to: Project / Research / Knowledge / Delete

## Active Projects (01-Projects)
- Project 1: {{progress}}
- Project 2: {{progress}}

## Agent Check-In (02-Agents)
- [ ] Review agent outputs from yesterday
- [ ] Update agent memory if context changed

## Knowledge Captured
- What did I learn today that belongs in 05-Knowledge-Base?

## Tomorrow
- Top 3 priorities...

Set this as a daily Obsidian template (Settings → Templates → Template folder location: Templates). Use the Ctrl+T shortcut or the Daily Notes plugin.

Connecting Your Vault to Local AI

Common Mistakes

Mistake 1: Everything in Inbox, Nothing Processed

Inbox is a temporary holding area. If it has more than 20 notes, your system is broken. Run the daily review.

Mistake 2: No Knowledge Base Extraction

When a project ends, don't just archive it. Extract the reusable lessons into 05-Knowledge-Base/. That's how your vault compounds in value.

Mistake 3: Not Using Frontmatter

Every note should have YAML frontmatter (tags, date, status). This lets you query your vault with Obsidian's Dataview plugin or feed structured data to your AI agents.

Recommended Setup

• Vault location: Local SSD (not cloud-synced folder — causes conflicts)
• Sync: Git (private repo) for version history
• Plugins: Dataview, Templater, Daily Notes
• AI model: qwen2.5:14b for vault queries, qwen2.5-coder:14b for code agents
• Backup: Cron job with rsync to external drive nightly

What I Would Do

Start with the six folders and the daily review template. Don't over-engineer. Use the vault for two weeks before adding agents or prompt automation. The structure matters more than the tools — once your notes are organized, connecting AI is the easy part. The 02-Agents/_memory/ folder is where the magic happens: it's how your local AI stops forgetting everything between sessions.

Frequently Asked Questions

What are the best AI plugins for Obsidian?

Top plugins: Smart Connections (semantic search and chat), Text Generator (drafting and outlining), and Copilot (ChatGPT-style conversations). For local AI, use the Ollama integration plugin to connect directly to local models.

Can AI plugins search my entire vault?

Yes, plugins like Smart Connections create embeddings of all notes for semantic search. Ask 'what have I written about productivity?' and get relevant results. With local models, this happens entirely offline.

Is my data private with AI plugins?

With Ollama-based plugins, data never leaves your computer. Cloud API plugins (OpenAI, Anthropic) send note content to servers. For privacy, use Ollama-based plugins and avoid cloud APIs for sensitive notes.

Can I use AI features offline?

Yes, connect Obsidian plugins to Ollama locally for full AI features without internet. Install Ollama, pull a model, configure the plugin to localhost:11434. Chat, summarization, and semantic search all work offline.

How long does setup take?

About 15-20 minutes: install Ollama and pull a model (5 min), install a plugin (2 min), configure to local Ollama (5 min). Indexing a large vault (1,000+ notes) may take an additional 10-30 minutes.

Last Updated: July 1, 2026 — Vault structure tested with Ollama, Open WebUI, and Goose on Ubuntu 26.04.