Obsidian AI plugins or a notes app with AI built in?
If you want to keep Obsidian and pick your own AI model, cloud or local, a plugin such as Smart Connections or Copilot is the way, and you decide where each request goes. If you want AI that works on the first launch, offline, without choosing a provider, managing a key or running a model server, a notes app with the AI built in does that. ThinkE is one, and it opens your Obsidian vault as it is, so you do not have to choose between them: both apps can open the same folder.
What Smart Connections and Copilot do
Smart Connections shows the notes that are related in meaning to the one you are writing, without a search query or existing links, and gives you a Lookup view to search the whole vault by meaning. Its README says a local embedding model does this with "zero setup" and "no API key". Chatting with your notes is a separate plugin, Smart Chat; connecting it to API models, cloud or local, is part of the Pro plugins.
Copilot (version 4) is built around an Agent for multi-step work, with Quick Chat for short questions, semantic search over the vault and inline commands. It runs on a model you provide (your own key for a cloud provider, or a local model), or on models Copilot hosts for paying users.
What they send, and where
Both plugins are careful about this, and both say it plainly:
- Smart Connections. Embeddings are made on your computer by default, and "Pro access does not upload notes by itself." A cloud embedding or ranking provider you configure "can receive eligible source text", and chat "may send prompts and returned sources to their provider". You can preview and edit the context before it is sent.
- Copilot. "Messages and note context go to the LLM provider, local endpoint, or CLI agent you configure." With your own key, requests go from your device to that provider, and Copilot's makers (Brevilabs) never see them. With a local model such as Ollama or LM Studio, requests stay on your machine. With Copilot's hosted models, requests pass through Brevilabs' backend in the United States to a model provider there; its privacy page says prompts and notes are not kept after the request and not used to train any model. "Hosted models and cloud-backed features require network access."
So with either plugin, the privacy of a note is the privacy of the model you point it at. A local model keeps everything at home, and setting one up is your job.
A vault ThinkE opens as it is
ThinkE reads your Obsidian vault in place: the same folder, the same Markdown files, nothing imported and nothing to export. What it adds is ordinary files beside yours.
- The AI comes with the app. It runs on your computer, Mac or Windows, and works offline once installed. A computer with 8 GB of memory gets the Lite AI; one with 24 GB, or a graphics card of 16 GB, gets the Standard AI. There is no provider to pick and no key.
- Search by meaning on every computer. Search finds your words first; when no note has them, Similar shows up to three notes close in meaning. It works even on a computer too small for the AI.
- Diagrams and mind maps. Ask the assistant to draw a flow, or choose Mind Map or Diagram of This Note from a note's … menu. A note with two or more headings gets its mind map instantly, by code.
- Summaries of documents. Choose Summarise on a PDF or Word document, and the summary is written into
a
.note.mdfile beside it. - Links survive renames. Renaming or moving a note in ThinkE updates the wikilinks and Markdown links that point to it.
ThinkE goes online to download its AI models, fetch the web pages you choose to save, check for updates, confirm your licence and send crash reports you can turn off, never with your notes.
What you give up: the models that run on a laptop are smaller than the largest cloud models, ThinkE has no plugin ecosystem of its own, and it has no phone app. Obsidian keeps all of those, and it still opens the same folder.
Side by side
| Compared | Obsidian + Smart Connections | Obsidian + Copilot | ThinkE |
|---|---|---|---|
| Setup | Install the plugin; related notes need no key. Chat is the separate Smart Chat plugin | Install the plugin, then add your own key, run a local model, or subscribe for hosted models | Install the app and open your vault; the AI models download on the first launch |
| Offline | Related notes, yes. Chat depends on the model you connect | With a local model. Hosted models need the network | Yes, once installed |
| Search by meaning | Yes: Connections and Lookup views | Yes, on the free plan | Yes, on every computer, even without the AI |
| Diagrams | Not among its listed features | Not among its listed features | Diagrams and mind maps, drawn by the AI or from a note's headings |
| Price | Free core; all-access Pro plugins $30 a month or $299 a year | Free with your own key or a local model; Plus $14.99 a month or $139.99 a year | $99 once, with a year of updates |
Plugin prices are each vendor's own, read on the date below; check their pages before you decide. ThinkE keeps working as it is after its year of updates. Try it free for 7 days with no card; you have 30 days to ask for your money back.
Sources
Read on October 13, 2026.