Obsidian AI: Built-In Features, Best Plugins, and Private Setup

As of October 2026, Obsidian has no built-in AI assistant inside the note app. “Obsidian AI” means community plugins such as Copilot, Smart Connections, Text Generator, and Local GPT, which connect your vault to a cloud model through an API key or to a local model through Ollama. The one official AI feature is Interpreter, part of the Obsidian Web Clipper.

That distinction matters more than any feature list. Obsidian itself stores notes privately on your device as plain Markdown files, and its privacy policy says local data is never sent to its servers. The moment you add an AI plugin, that changes: your notes can travel to whichever model provider you choose. This guide explains which plugins are worth a look, how to install one, what it costs, and how to keep a private vault private.

Related reading: Ollama tutorial, Zotero AI, and AI notes generator. Key terms used in this guide: on-device AI, prompt injection, and API key.

Does Obsidian have built-in AI?

No. The Obsidian changelog lists desktop releases up to version 1.14 in early October 2026, and none of the recent entries adds an AI assistant to the app. The Obsidian homepage does not mention AI either. Every AI feature you see in Obsidian videos comes from a community plugin or an outside tool.

There is one partial exception. The free Obsidian Web Clipper browser extension has a feature called Interpreter, which uses a language model to summarize, extract, or translate a web page while you clip it into your vault. It works with providers such as Anthropic, Google Gemini, OpenAI, and Ollama. You bring your own API key, except with Ollama, which needs none. Obsidian states that it does not gather or store any data about these requests.

So when someone says “Obsidian AI,” they usually mean one of three things:

• A chat panel that answers questions about your notes (Copilot, Smart Chat)
• Semantic search that finds related notes by meaning (Smart Connections)
• Writing actions on selected text, such as summarize or fix grammar (Text Generator, Local GPT)

Why use AI in Obsidian?

A vault grows faster than your memory. After a few hundred notes, it becomes hard to remember what you already wrote, and plain keyword search only finds exact words. AI tools help with that problem in a few practical ways.

Finding forgotten notes. Semantic search looks for meaning rather than exact matches. A note titled “Budget review March” can show up when you are writing about “household spending,” even though the words differ.

Summarizing long material. Meeting notes, book highlights, and clipped articles can be condensed into a few bullet points you can scan later.

Drafting and editing. You can ask for an outline, a cleaner paragraph, or a list of action items pulled from a messy note.

Asking questions across the vault. Chat plugins let you ask, “What did I decide about the garden layout?” and get an answer that points back to your own notes.

Here is a concrete example. Imagine (a hypothetical case) a retired teacher who keeps 600 notes on family history. Names and dates are scattered across interview notes, scanned letters, and research logs. With semantic search, she opens a note about one grandparent and sees five related notes she had forgotten. With a chat plugin, she asks for a timeline of one branch of the family and gets a draft she then checks against the original notes. The AI saves searching time; she still does the checking.

That last step is the real benefit and the real limit. AI in Obsidian works best as a research assistant that points you to your own material, not as a source of truth.

Popular Obsidian AI plugins compared

The table below covers five widely used plugins. Versions and dates come from each project’s GitHub page, checked on 13 October 2026. Plugins change quickly, so check the page again before you install.

Copilot is the most feature-rich option. Its version 4 brings agents such as opencode, Claude Code, and Codex into Obsidian, and it won the language models category in the 2024 Obsidian Gems of the Year. The plugin is open source and works without a paid plan when you use your own key, your own Claude or ChatGPT agent login, or a local model. Paid plans add hosted models from its maker, Brevilabs. Two details stand out on its GitHub page: Agent mode is desktop only, and keys you add are stored in the device’s Obsidian Keychain rather than in the vault’s settings file.

Smart Connections is the easiest start. Its page says it ships with a local embedding model, so you install it, enable it, and it starts indexing with no API key. It shows related notes in a side panel as you write. Chat moved to a separate Smart Chat plugin, and API-based model connections, cloud and local, now sit in Smart Chat Pro, which is offered to project supporters.

Text Generator suits people who like reusable templates, such as “summarize this note in five bullets.” It is free and open source, and its settings support several cloud services and local models.

AI Providers and Local GPT work as a pair from the same developer. AI Providers stores your provider settings once, and Local GPT adds actions like continue writing, summarize text, fix spelling and grammar, and find action items. This is the most direct route to a fully local setup with Ollama.

Getting started with Obsidian AI

Installing any AI plugin follows the same pattern. The main choice is where the model runs: on a provider’s servers, or on your own computer.

Install a plugin step by step

Obsidian starts in Restricted mode, which blocks all third-party code. According to Obsidian’s plugin security page, you turn it off like this:

1. Open Settings.
2. In the side menu, select Community plugins.
3. Select Turn on community plugins. Do this only for plugins whose authors you trust; see the security section below.
4. Select Browse, search for the plugin by name, then select Install and Enable.
5. Open the plugin’s own settings page and choose a model provider.

If you pick a cloud provider, you create an API key in that provider’s developer console and paste it into the plugin. Treat that key like a password: do not paste it into a note, a shared document, or a chat window.

Local setup with Ollama for privacy

Ollama runs language models on your own computer. Its FAQ says that when you run locally, Ollama does not see your prompts or data. A simple path, based on the AI Providers instructions:

1. Install Ollama on your computer.
2. Download a model with a command such as ollama pull gemma2.
3. In Obsidian, install AI Providers, add a provider, and set the type to Ollama.
4. Select the refresh button and pick your model.
5. Install Local GPT and assign a hotkey to its action menu.

Two cautions. First, Ollama also offers cloud-hosted models, so check that the model you pick runs on your machine if privacy is the goal. The Ollama FAQ also describes a local only mode: set the environment variable OLLAMA_NO_CLOUD=1, or set "disable_ollama_cloud": true in ~/.ollama/server.json, then restart Ollama. Second, local models depend on your hardware. A small model on an older laptop will be slower and less accurate than a large cloud model.

Troubleshooting common issues

• Plugin missing from the list. Restricted mode is still on. Turn on community plugins and try again.
• Ollama connects but answers stall or fail. If streaming fails, add only the specific origin your plugin’s documentation names for Obsidian to OLLAMA_ORIGINS. Never use *, and never set OLLAMA_HOST to 0.0.0.0. The Ollama FAQ explains that only certain origins are allowed by default, and the Web Clipper needs its extension origin allowed explicitly.
• Answers ignore the end of a long note. Local models have a limited context window. Ollama’s FAQ gives a default of 4096 tokens, which you can raise with OLLAMA_CONTEXT_LENGTH. Shorter, focused notes also help.
• “Invalid key” or billing errors. Check that billing is set up in the provider’s developer console. API access can be billed separately from a chat app subscription.
• A plugin stops working after an update. Check its GitHub issues page, and confirm your Obsidian version meets the plugin’s minimum.

What to know before deciding: security, privacy, and cost

This is the section most guides skip, and it is the one that protects your notes.

Community plugins have full access

Obsidian is direct about this. Its plugin security page says it cannot reliably restrict plugins to specific permissions. Plugins inherit Obsidian’s access level, which means they can read files on your computer, connect to the internet, and install other programs. Obsidian scans every plugin version for vulnerabilities and malware and shows a safety scorecard on each plugin’s directory page, with manual reviews for popular, featured, and flagged plugins. For sensitive data, Obsidian recommends an independent security audit before use.

In practice, that means a few habits:

• Install plugins with many users, recent releases, and an active issues page, and check the safety scorecard on the plugin’s directory page.
• Keep the number of AI plugins small. Each one is more code with full access.
• Turn Restricted mode back on in a vault where you do not need plugins.

Where your notes go

With a cloud provider, the text you send leaves your computer and is handled under that provider’s terms. Copilot’s README states that in free use, messages and note context go to the provider or agent you configure, and that its hosted features send requests to Brevilabs.

Some content should never reach a cloud AI tool: passwords, ID or passport numbers, card numbers, health records, and anything about children. If your vault holds that material, keep it in a separate vault with no AI plugins, or use a fully local model.

AI agents add one more risk. When an agent reads web clips or outside documents, hidden text in that content can try to steer it, a problem called prompt injection. Give agents access only to the folders they need, and keep approval steps on for moving, renaming, or deleting files.

What API usage costs

Obsidian itself is free without limits, and most AI plugins are free to install. The cost comes from the model. Cloud APIs bill by tokens, which are small chunks of text, counted separately for input and output. As of 13 October 2026, Anthropic’s API pricing lists Claude Sonnet 5.5 at $2 per million input tokens and $10 per million output tokens, and Claude Haiku 5.5 at $0.10 and $0.50 for prompts up to 100,000 tokens. Verify current pricing on the official site before you rely on these numbers.

A worked example: one request that sends 10,000 tokens of notes and gets 1,000 tokens back would cost about 2 cents of input plus 1 cent of output on Sonnet 5.5, roughly 3 cents in total. Vault-wide chat and agents can send far more text per request, so set a monthly spending limit in your provider’s console if it offers one. Local models through Ollama have no per-token fee; the cost is your hardware and electricity.

Decision framework: choosing your setup

Use this checklist to pick a starting point. Answer each question in order and stop at the first clear match.

1. Do your notes include private or work-confidential material? Start local: Smart Connections for search, or AI Providers with Local GPT and Ollama for writing help.
2. Do you mainly want to find old notes? Smart Connections alone may be enough. No key, no cost.
3. Do you want to chat with your vault or run agents? Copilot, with your own key or a local model first. Add a paid plan only if you need its hosted features.
4. Do you repeat the same writing tasks? Text Generator templates fit that pattern.
5. Do you mostly clip web articles? Try Interpreter in the Web Clipper before adding any plugin.

Three hypothetical scenarios show how this plays out:

• A small business owner keeps client notes in Obsidian. She runs Smart Connections in that vault and uses a cloud chat plugin only in a second vault of general research notes.
• A hobby writer wants help outlining chapters. He installs Copilot with an API key, sets a monthly cap, and starts with a cheaper model for brainstorming.
• A privacy-minded retiree with an older laptop tries Ollama with a small model, finds it slow, and switches to Smart Connections for search, which needs no separate AI app.

Common mistakes to avoid

• Pointing an AI plugin at the whole vault on day one instead of one test folder
• Keeping API keys inside notes
• Trusting AI summaries without opening the source note
• Installing several AI plugins that do the same job

Migrating existing notes. You do not need to restructure your vault. A safe path is to copy one folder into a new test vault, try a plugin there for a week, then enable it in your main vault. Clear note titles and short notes improve results more than any setting.

Next step

Pick one small folder of notes, choose a setup from the checklist above, and try a single plugin for a week. Write down what saved you time and what you had to double-check. If you want a guided way to learn how prompts, models, and AI privacy work before you connect your notes, Explore Coursiv AI lessons for short, step-by-step practice at your own pace.

Related guides: Notion AI for students and best AI note takers.

FAQ