You have a few hundred notes in Obsidian and you want your AI to actually use them. The community plugin store lists dozens of options with overlapping names, and half of them wrap the same OpenAI call behind a different settings panel. This guide names the seven that do real work, tells you exactly what each one does, and flags the one job none of them handle: keeping the context current after you install it.
Every plugin below is a real, installable Obsidian community plugin. I have grouped them by the job you are hiring them for: chat over your vault, semantic linking, generation, and search. Pick the one that matches the job, not the one with the most GitHub stars.
Key takeaways
- The best Obsidian AI plugins split into four jobs: chat with your notes, surface related notes, generate text, and search semantically. Match the plugin to the job.
- Copilot for Obsidian and Smart Connections are the two most-installed, and they solve different problems. Most people end up running both.
- Almost every plugin needs your own OpenAI, Anthropic, or local (Ollama) API key. Budget for tokens or run a local model.
- Local-first options exist. Smart Connections and Copilot both support local embeddings and Ollama, so your notes never have to leave the machine.
- A plugin reads your vault at query time. It does not maintain a distilled, current model of your context, which is why AI answers drift as your notes pile up.
---
How to pick an Obsidian AI plugin
Before the list, get the shape of the decision right. Every AI plugin for Obsidian answers one of four questions: can I chat with my whole vault, can the app surface related notes as I write, can it draft or expand text in place, and can it search by meaning instead of exact keywords.
Two practical constraints decide which one you keep installed. First, the model. Most of these plugins are bring-your-own-key: you paste an OpenAI or Anthropic key, or point them at a local model through Ollama. If you want everything to stay on your machine, filter for local support up front. Second, the index. Semantic plugins build an embedding index of your vault, and on a large vault that first index takes minutes and re-runs as notes change, so watch the cost and the CPU.
Here is the shortlist mapped to the job each one does best.
| Plugin | Primary job | Bring your own key | Local model support | Best for |
|---|---|---|---|---|
| Copilot for Obsidian | Chat over your vault | Yes | Yes (Ollama) | Asking questions across all notes |
| Smart Connections | Related notes + chat | Yes | Yes (local embeddings) | Surfacing links you forgot you made |
| Text Generator | Draft and expand text | Yes | Partial | Templated writing inside a note |
| Obsidian AI (various) | Inline assistant | Yes | Partial | Quick rewrites and summaries |
| Smart Second Brain | Local RAG chat | Yes | Yes (Ollama) | Private chat with no cloud calls |
| Companion | Autocomplete | Yes | Partial | Ghostwriter-style completions |
| Semantic Search | Meaning-based search | Yes | Yes | Finding notes by concept |
Now the detail on each.
---
1. Copilot for Obsidian
Copilot is the plugin most people mean when they say "Obsidian AI assistant." It adds a chat panel to your right sidebar and can answer questions using the note you are in, a set of notes you select, or your entire vault through its Vault QA mode. That last mode is the one worth installing for: it embeds your vault and does retrieval-augmented chat, so you can ask "what did I decide about the pricing model" and get an answer pulled from the actual notes.
It is bring-your-own-key and model-agnostic. Wire it to OpenAI, Anthropic, Google, or a local model through Ollama, and you can run the whole thing offline with local embeddings. The paid Copilot Plus tier adds hosted models, but the free plugin with your own key covers the core chat job.
Where it strains: Vault QA is only as good as its index, and on a big vault the retrieval sometimes misses the note you were sure it would find. It is a search-and-summarize tool, not a curated memory, so answers reflect whatever raw text happens to match.
2. Smart Connections
Smart Connections does the thing Obsidian's own graph view promises but rarely delivers: it shows you notes that are semantically related to the one you have open, ranked by similarity, in a sidebar that updates as you move around. These are connections you never made by hand, surfaced from meaning rather than from [[wikilinks]]. For anyone with a large vault of orphan notes, this is the single most useful AI feature in Obsidian.
It also ships a chat mode (Smart Chat) so you can query the vault conversationally. Crucially, it runs local embeddings by default, so the indexing that powers the "related notes" panel happens on your machine and your notes are not shipped to a third party unless you opt into a cloud model for chat.
Where it strains: the first full index of a several-thousand-note vault is slow, and the similarity results are only as meaningful as your notes are distinct. Near-duplicate notes produce near-duplicate suggestions.
A plugin can read every note you have. It still has no idea which ones are still true.
3. Text Generator
Text Generator is built for writing, not chatting. You define a prompt template with placeholders, invoke it inside a note, and it generates or expands text in place. Think of it as a programmable writing macro, useful for anyone with a repeatable structure, like turning bullet notes into a draft or expanding an outline heading into a paragraph. It is bring-your-own-key and works with the major providers, and the template system is the differentiator: encode a document structure once and reuse it.
Where it strains: it generates from the prompt and the current note, not your whole vault, so it does not "know" your other notes unless you paste them in. It is a generation tool, not a memory tool.
4. Obsidian AI assistant plugins (inline)
There is a cluster of plugins that add an inline AI assistant: highlight text, run a command, and get a rewrite, summary, translation, or continuation. They are convenience wrappers around a chat model, and the good ones let you define your own command prompts. These earn their place by removing friction. You are already in the note and do not want to open a chat panel and copy text back and forth, so for quick rewrites an inline assistant is faster than any dedicated chat UI.
Where it strains: same ceiling as Text Generator. The model sees the selected text and maybe the current note, nothing more. It cannot reason across your vault.
5. Smart Second Brain
Smart Second Brain is built around private, local retrieval-augmented chat. It connects to Ollama to run models locally and indexes your vault so you can chat with it without any cloud call. For a privacy-first setup where nothing leaves the machine, this is the most complete option on the list, and the plugin for anyone who wants the Copilot Vault QA experience but refuses to send notes to a hosted API. The tradeoff is the usual one: answer quality depends on the local model you can run, and a laptop-sized model is weaker than a frontier hosted one.
Where it strains: local models and local indexing are heavier on hardware, and setup (installing Ollama, pulling a model) is a real step, not a one-click install.
6. Companion (autocomplete)
Companion adds ghostwriter-style autocomplete to Obsidian. As you type, it suggests a completion in gray text that you accept with a keystroke, the same interaction pattern as a coding autocomplete tool. If you draft long-form prose in Obsidian and want momentum, it keeps you moving.
It supports multiple providers and lets you tune how aggressive the suggestions are. Bring your own key.
Where it strains: autocomplete is a style and momentum aid, not a research aid. It predicts plausible next words; it does not retrieve facts from your notes. Treat its output as a first draft you will correct.
7. Semantic Search
Semantic Search plugins let you search your vault by meaning. Instead of matching the literal string "renewal terms," a semantic query finds the note about "contract expiry" even though the words differ. This closes the biggest gap in Obsidian's built-in search, which is keyword-only and misses anything you phrased differently six months ago.
It builds an embedding index of your notes and ranks by similarity to your query. Several implementations exist; the ones worth keeping run local embeddings so the index stays on your machine.
Where it strains: it is search, not synthesis. It hands you the matching notes; you still read and combine them yourself.
---
What every one of these plugins cannot do
Look back at the list. Each plugin reads your raw notes at the moment you query, then throws that context away. None of them build a durable, distilled model of what you know, and none keep it current when a fact changes.
That gap matters more than any feature comparison. Say you wrote a note six months ago that a project used Postgres, then moved it to a different database and wrote a new note. Ask any vault-chat plugin about the stack and it surfaces both, because both still exist as raw text. It has no notion that the second supersedes the first, so you get a confident answer built on a fact that is no longer true. Garbage in, wrong out.
This is not a plugin bug. Obsidian was built as a note editor, and these plugins are search-and-generate layers on top of it. Maintaining a second brain that stays true over time, pruning stale notes, marking what changed, tagging entities, is a full-time job that no amount of plugin polish removes.
That is the specific problem Locul is built for. It is a second brain that builds itself from what you already do and keeps itself current, then serves that context to your AI over MCP. Two things make it different from a plugin:
- It reads your data where it lives. It ingests your local markdown notes, including your
#tags,[[wikilinks]], and backlinks, plus PDFs and dictation, without forcing everything into one vault. You keep editing and viewing your brain the way you already do. - It stays current on its own. It distills notes into memories tagged as fact, preference, decision, event, or insight, each with a confidence score. When a fact changes, the old one is marked superseded and the history is preserved. That supersedence step is the exact thing plugins skip, and it is why AI answers stop drifting.
So the honest recommendation: install a plugin for the job it does well. Use Smart Connections to resurface forgotten notes, Copilot to chat, Text Generator to draft. Then, if you actually depend on your AI reasoning over your knowledge, add a layer whose whole job is keeping that knowledge true. A current, specialized brain is where the real quality jump in AI output comes from, and it is the one thing the plugin store does not sell.
If your capture happens by voice rather than typing, the notes those plugins read have to exist first. Dictation tools like Contextli turn spoken thought into clean notes that flow into your vault and into Locul.
---
FAQ
What is the best Obsidian AI plugin overall?
There is no single winner, because the plugins do different jobs. For chatting with your whole vault, Copilot for Obsidian is the most common pick. For surfacing related notes automatically, Smart Connections is the standout. Most people who lean on AI in Obsidian run both. If privacy is the priority, Smart Second Brain with Ollama keeps everything local.
Do Obsidian AI plugins need an API key?
Almost all of them do. Copilot, Smart Connections, Text Generator, and the inline assistants are bring-your-own-key, meaning you supply an OpenAI, Anthropic, or Google key and pay that provider for tokens. The local-friendly ones (Smart Connections, Copilot, Smart Second Brain) can instead point at a model running locally through Ollama, which has no per-token cost but needs capable hardware.
Are Obsidian AI plugins private?
It depends on the model, not the plugin. If a plugin sends your note text to a hosted API like OpenAI, that text leaves your machine. If you configure it to use local embeddings and a local model through Ollama, everything stays on your device. Smart Connections runs local embeddings by default, and Smart Second Brain is built around fully local operation. Check each plugin's model settings before assuming privacy.
What is the difference between Copilot and Smart Connections?
Copilot is chat-first: you ask questions and it answers using retrieval over your vault. Smart Connections is connection-first: it continuously shows notes related to whatever you are viewing, and adds chat as a secondary mode. Copilot is for interrogating your notes; Smart Connections is for discovering links between them. They complement each other rather than compete.
Can Obsidian plugins keep my AI context up to date automatically?
Not really. A plugin reads your raw notes at query time, so if a note is out of date, it will happily use it. It has no mechanism to notice that a newer note supersedes an older one. Keeping context current, marking old facts as superseded and distilling notes into clean memory, is a separate job from search and generation, and it is the gap a purpose-built second brain such as Locul fills.
Will an AI plugin slow down a large Obsidian vault?
The semantic plugins can, during indexing. Smart Connections, Copilot's Vault QA, and semantic search all build an embedding index, and the first full pass on a vault of several thousand notes takes minutes and some CPU. After that they re-index incrementally, which is lighter. Chat and inline-assistant features that do not index have negligible impact on vault performance.
---
Plugins get you a chat box and a related-notes panel. What they never give you is a context that stays true as your notes pile up and your facts change. If you want your AI to work from your real, current knowledge instead of a snapshot, start with Locul free: 500 active memories, local AI, no credit card. Read how to give your AI a memory that lasts for the mechanism, or see what a Memory Pack is if you want to share a curated brain. When you are ready to compare tiers, the pricing page lays out Free, Pro, and Studio.
FAQ
