Obsidian AI workflows knowledge graph productivity structure abstract

Obsidian AI Workflows: 5 Honest Patterns From a Solo Stack

by

Obsidian AI workflows clicked for me this month after two false starts in 2024 and 2025. The unlock wasn’t a new feature — it was the Smart Connections plugin reaching a state where the local embeddings actually surfaced the right note without me hunting for it. I migrated my client knowledge base off Notion AI into an Obsidian vault in early May 2026, ran it for four weeks beside my existing stack, and the patterns that survived the trial are the five I’m writing down here. One of them I dropped after a week. Two of them changed how I prep for client calls. The honest version follows — what worked, what didn’t, and where Notion AI still wins.

In this article

  • Why Obsidian AI workflows clicked this month
  • The 5 patterns from my first month
  • Where the workflow still loses to Notion AI
  • What I’m changing next month

Why Obsidian AI workflows clicked this month

Three things turned Obsidian AI workflows from “interesting idea” into “weekly habit” for me. The first is Smart Connections. The plugin uses local embeddings to surface notes related to whatever I’m writing — even when they share zero keywords — and the May 2026 build runs the embedding model on-device, so no vault content leaves the machine. That single property pulled the privacy-sensitive client work into Obsidian that I’d kept out of Notion AI for two years.

The second is Copilot for Obsidian, which adds a ChatGPT-style chat that grounds answers in the vault through Vault Q&A. Where Smart Connections finds related notes, Copilot answers questions across them. The third is the plugin ecosystem maturing past the “weekend project” feel — there are now stable, openly maintained AI plugins that don’t break with each Obsidian point release. That combination — local-first embeddings, vault-grounded chat, plugin stability — is what made Obsidian AI workflows worth the migration time. The reason it landed in May 2026 and not earlier is that all three only stabilized in the last quarter.

The 5 patterns from my first month

Five patterns earned weekly use in the first month of Obsidian AI workflows. They’re listed in order of how much time they save.

  • Pattern 1 — Smart Connections as a research backstop. Before drafting any client brief I let Smart Connections suggest five related notes from past projects. About a third of the time it surfaces something I’d forgotten — a competitor I’d analyzed eighteen months ago, a positioning angle from a different industry that maps cleanly to the current client.
  • Pattern 2 — Copilot Vault Q&A for client meeting prep. The morning before a client call I ask Copilot “what have I noted about [client name] this year?” and get a single answer grounded in every meeting note, brief, and Slack export I’ve dumped in. The same query in Notion AI took three searches and a manual stitch.
  • Pattern 3 — Local embeddings for client-confidential vaults. Each client has a separate vault with embeddings cached locally. Smart Connections runs without any API calls for these vaults, which closes the data-routing question that kept this work out of cloud tools.
  • Pattern 4 — Templater plus AI for repeatable brief structures. A Templater note injects the standard 6-section client brief skeleton, then a Copilot prompt fills the first-draft text from a tagged source folder. The structure stays consistent across clients without me copy-pasting from old briefs.
  • Pattern 5 — Daily note review with AI summary. End-of-day, Copilot summarizes the day’s notes into three lines I paste into my weekly review. This is the smallest win but the most consistent — five minutes saved every day adds up faster than I expected.

The pattern I dropped after a week was using AI to write new note content from scratch in the vault. The drafts felt generic and disconnected from my actual thinking — the value of Obsidian AI workflows is augmentation, not generation.

“Smart Connections uses AI embeddings to find notes related to what you are currently writing — even if they share zero keywords in common.” That’s the line that explains why this loop works where Notion AI’s search doesn’t.

Where the workflow still loses to Notion AI

Obsidian AI workflows lose to Notion AI in three places, and the gaps are honest enough to acknowledge upfront. The first is collaboration. Obsidian is single-player by default; collaborative editing requires Sync or a hosted Git repo with manual conflict resolution. For solo work this doesn’t matter — for client-facing shared docs, Notion AI’s collaborative surface still wins and I haven’t tried to move that work into Obsidian.

The second loss is database-style structure. Notion’s tables, filters, and rollups beat anything I’ve cobbled together in Obsidian with Dataview. Project tracking and client CRM-lite stayed in Notion for that reason. The third loss is mobile capture. The Obsidian mobile app works, but quick capture from my phone is still smoother in Notion. About 20% of my note creation happens on mobile, so this gap is real.

The cleanest rule I’ve landed on after a month of running both: Obsidian AI workflows handle thinking, drafting, and reference work; Notion handles structured project state and shared client docs. Splitting them by job rather than by tool kept the migration sane.

What I’m changing next month

After four weeks of Obsidian AI workflows, three changes are queued for June. First, I’m wiring Hermes 4 70B (running locally via Ollama, as I covered in the local LLM piece I drafted this week) into Copilot as the answering model for sensitive vaults. That removes the cloud dependency entirely for the client-confidential workflows. Second, I’m building a per-client weekly digest template that runs the Copilot Vault Q&A prompt automatically on Friday afternoons. Third, I’m migrating the last 14 months of meeting notes from Notion into a single archive vault so Smart Connections has a deeper context pool to draw from.

For me, Obsidian AI workflows are the rare tool change in 2026 that earned its slot without displacing what was already working. Notion stayed for the jobs it does well. Obsidian took over the jobs Notion was always slightly wrong for. The plugins are still maturing — Smart Connections in particular feels like it’s two releases from being polished — but the underlying loop is already worth the setup time. That’s the call I’m making after a month of real use, with the caveat that I’ll revisit it before Q4 if either plugin slows down.

The broader pattern is also worth naming. Three of the five tool changes I’ve made this quarter — Ollama, Hermes 4, and now Obsidian — share a property: they’re locally-run, open-weight, or both. None of those choices were ideological. Each landed because the cloud-only version was missing something specific the local version could deliver. For a solo consultant in 2026 that pattern is becoming the default rather than the exception, and Obsidian sits squarely in the middle of it.

Sources

AI-assisted research and drafting. Reviewed and published by ToolMint.