No. Obsidian's desktop app has no built-in AI assistant. Because your notes are plain Markdown files there are three ways to add one, and here is our own proposal for choosing: ask where the assistant runs, whether Obsidian must be open, and whether it can change your notes. In our tests of the folder-reader route, a search for a note's exact title returned that note first in 40 of 40 cases.
Obsidian's desktop app does not include a general AI assistant for chatting with, rewriting or summarizing your vault. Its one official AI feature is Interpreter, part of the separate Web Clipper browser extension. It processes a web page before you save it. It is not a vault-wide assistant, and you pick and configure the model it uses.
What Obsidian does have is a vault that is a folder of Markdown files. Anything that can read a folder can add AI to it, which is why there are several different ways to do it.
1. A plugin inside Obsidian. Two names come up most. Smart Connections finds related notes and offers semantic search; one independent guide describes it as free and offline, and says it does not upload your notes. Copilot adds editing help, a chat mode and question answering across the vault; the same guide says it needs an API key, costs a small amount, and sends note content to cloud language models. Community plugins in general need Restricted mode turned off, can reach your files and the internet, and do not update on their own. The assistant lives inside Obsidian, so it only works there.
2. An MCP server running inside Obsidian. The Local REST API plugin now includes an MCP server at a local address, protected by a bearer token. Its page gives setup steps for Claude Code, Claude Desktop (through the mcp-remote bridge), Cursor and any client that speaks Streamable HTTP. The tools cover search, reading, writing and deleting notes; deletes go to the trash by default, and the page does not mention a read-only setting. Because it is a plugin, Obsidian has to be running. This is the closest thing to an assistant working inside your vault, which also means any client holding the key can change it.
3. A memory outside Obsidian that reads the folder. ModelBrain is a local memory that any MCP assistant can query. You add the vault folder once and it reads the Markdown files where they sit, keeps its own index, and answers with the note each answer came from. There is no plugin and Obsidian does not need to be open.
The Obsidian guide has the full detail, including what is read and what is skipped.
Want AI help writing and editing while you are inside Obsidian? Use a plugin, and check where its model runs. Want an assistant to create and edit notes through a tool? The MCP plugin does that, so decide whether you are comfortable giving a client write access. Want the same notes answerable from Claude, ChatGPT and Cursor, with no plugin and Obsidian closed? That is the folder reader. They do not conflict: all of them read the same Markdown files.
These figures come from the tests behind the Obsidian guide, run on a sample vault and, where noted, with the real model. A search for a note's exact title returned that note first in 40 of 40 cases. On a 300-note vault a saved note became searchable in about 2.1 seconds. The first pass over a new vault is slower: 500 notes took 46 seconds with the real model, and a large vault takes longer. Re-reading notes that have not changed is near instant.
The limits are plain. Sync is not instant for edits and deletions made while it is running. ModelBrain keeps no earlier versions, so an assistant sees what a note says now. Images, video and scanned PDFs are skipped. The link that opens a note in Obsidian is checked on macOS only, so it is not claimed on Windows. Moving your memory between computers is a separate, optional paid feature. ChatGPT needs the free account and a plan that supports custom plugins. ModelBrain is independent and is not affiliated with or endorsed by Obsidian.
For how recall quality is measured, see the recall test results, and for deleting what it learned, what a forget call actually guarantees. If you want the wider setup across apps, read how to give Claude, ChatGPT and Cursor the same persistent memory.
What exactly is Obsidian AI? There is no single product by that name. It means any AI you attach to an Obsidian vault: a plugin, an MCP connection, or a separate tool that reads the folder.
Is it free? Obsidian itself has no built-in AI to charge for. The plugins vary: Smart Connections is described as free and offline, while Copilot needs an API key and costs a small amount per use. Local use of ModelBrain needs no account; its paid plan covers sync and backup between computers.
Does Obsidian send my notes to the cloud? Obsidian describes itself as storing notes privately on your device. What leaves depends on the AI you add and which model it calls, so check that before you connect anything.
Last checked October 11, 2026. Plugin features change often, so confirm a detail on the plugin's own page before relying on it.