Guide · Updated May 2026

The complete guide to AI in Obsidian (2026).

Searches for "obsidian ai" are up 662% year-over-year. People want AI in their notes — but Obsidian wasn't designed for it. Here's every working path in 2026: the four major plugins, what they actually cost, the setup walkthroughs, and when you should stop assembling your own stack and pick a tool that ships AI native.

Why people want AI in Obsidian in the first place.

Three jobs, mostly:

1. Semantic search — "what have I written about X?" answered against your whole vault, not just keyword matches.
2. Chat with citations — ask a question, get an answer that cites the notes the model read.
3. Inline editing — select a paragraph, hit a shortcut, get a rewrite / summary / translation right in your editor.

Obsidian's core product is Markdown linking and graph visualization. The team has been deliberate about not shipping AI as a built-in feature — they leave it to the plugin ecosystem. That works, but it puts the assembly tax on you.

Plugin #1: Smart Connections

What it does: generates local embeddings of every note in your vault, surfaces "related notes" in the sidebar, and lets you chat with your vault using your own OpenAI / Anthropic / Gemini key.

Setup:

1. Settings → Community plugins → Browse → search "Smart Connections" → install + enable.
2. Open the plugin's settings, pick an embedding model. The free local option uses a quantized model that runs on your CPU.
3. Optional: paste an OpenAI or Anthropic key for higher-quality embeddings and chat.
4. Wait. First-time indexing of a 1000-note vault takes 3-15 minutes depending on your machine.

What works: the local embedding option means semantic search runs with zero per-query cost. The "related notes" sidebar is good. Vault chat with citations is solid.

What doesn't: indexing is slow on first run and re-indexes when the plugin updates. The UI lives in a sidebar pane that interrupts your writing flow. Citation chips don't always link cleanly to the source heading. And if the plugin author takes a month off, you're stuck on the old version.

Real cost: $0 if you use local embeddings + your existing Claude/ChatGPT subscription. ~$0.50-$3/day if you use OpenAI's hosted embeddings + GPT-4o for chat.

Plugin #2: Copilot for Obsidian

What it does: chat sidebar that can search your vault and answer with citations. Closer to ChatGPT-in-your-notes than Smart Connections, which is more retrieval-focused.

Setup:

1. Install from Community plugins.
2. Configure your LLM provider — OpenAI, Anthropic, Azure, OpenRouter, or local via LM Studio / Ollama.
3. Pick a "QA mode" if you want vault-aware answers (otherwise it's just a chat panel that doesn't see your notes).

What works: the chat UI is polished, supports image uploads (with vision-capable models), and the QA mode searches your vault before answering.

What doesn't: overlaps heavily with Smart Connections. If you run both, you have two semantic indexes of your vault eating disk. The QA mode's retrieval quality depends on Obsidian's built-in search, which isn't designed for AI-grade retrieval.

Real cost: $20-30/month at typical usage with Claude or GPT-4o.

Plugin #3: Text Generator

What it does: the original Obsidian AI plugin. Inline command-style generation — write a prompt as Markdown, hit a shortcut, the response replaces or appends to your note.

Setup:

1. Install from Community plugins.
2. Configure provider + model.
3. Optionally install template packs (community-built prompts for summarization, expansion, translation, etc.).

What works: the template system is genuinely powerful — you can define complex prompts as reusable templates and chain them. Inline command UX is closer to what you'd want from a "writing assistant" than the sidebar chat plugins.

What doesn't: the template syntax has a learning curve. Default templates feel dated. The plugin shows its age in places.

Real cost: usage-based with your own API key, typically $5-15/month for moderate use.

Plugin #4: BMO Chatbot + Ollama

What it does: chat with a local LLM running on your machine via Ollama. Zero cloud, zero API cost, but you bring the GPU.

Setup:

1. Install Ollama on your Mac.
2. ollama pull llama3.1:8b (or any model that fits in your RAM).
3. Install BMO Chatbot from Obsidian Community plugins.
4. Point BMO at http://localhost:11434.

What works: truly free, truly private, runs offline. Great for sensitive vaults (legal notes, health records, journaling).

What doesn't: local models are noticeably weaker than GPT-4o or Claude. 8B-parameter models can't follow complex prompts the way a frontier model does. If your Mac has < 16GB RAM, you'll struggle to run anything useful. Streaming is slower than cloud.

Real cost: $0 once you've paid for the hardware.

The hidden cost: plugin maintenance.

Every Obsidian update has the potential to break a plugin. Most plugins are maintained by one or two volunteers. When Obsidian ships a breaking change to their plugin API (it happens twice a year), some plugins update within days, others within weeks, some never. Your job is to track which.

If you run four AI plugins, that's four communities to monitor, four GitHub issue trackers, four upgrade paths. None of this matters until one breaks the week you have a deadline.

The hidden tax on the Obsidian + plugins path is that you become the systems integrator. For some people that's fun. For most, it's a job they didn't sign up for.

When to skip plugins entirely.

If you're going to write AI into your Markdown workflow seriously, there's a case for skipping the Obsidian + plugins approach and using a tool built around AI from day one. Eyrie is one option — a native Mac Markdown editor that ships AI as a core feature rather than a plugin.

The differences that matter:

• One install for chat with citations, inline ⌘E commands, semantic search, and integrations with Claude Desktop / Cursor / Windsurf.
• One developer maintaining the AI features alongside the editor — when one updates, both update together.
• BYO API key means you pay only for the AI calls, no markup, no monthly subscription on top.
• Or skip the key entirely by connecting Eyrie to your existing Claude Desktop subscription via MCP. Eyrie acts as the bridge between your notes and the AI you already pay for.

The trade-off: Eyrie is Mac-only, the editor is younger than Obsidian, and there's no Obsidian-style plugin marketplace. If those matter to you, stay with Obsidian + plugins.

FAQ