Obsidian AI Plugins: 5 Options for Search, Writing, and Agents

Obsidian AI plugins fall into three useful groups. Some find related notes, some help you write, and some let an agent read or change files in your vault. The best choice depends on which of those jobs you actually need.

I use Obsidian as the working memory behind my own AI operating system. That makes permission scope, local processing, and reliable retrieval more important to me than adding a chat box because it looks clever.

This comparison covers five current community plugins with different boundaries. I have not treated download counts as proof of quality. I compared the jobs they claim to do, how they handle models, and what access they need.

Which Obsidian AI plugins fit each job?

Pick the plugin that solves your main bottleneck. Semantic search is a different problem from inline editing, and both are smaller problems than giving an agent write access to an entire vault.

Copilot is the broad all-in-one option

Copilot V4 is the most complete option in this group. Its Agent view can inspect notes, use tools, create files, and continue across several turns with permissions you control.

It can connect OpenCode, Claude Code, or Codex on desktop. It also supports hosted models, your own provider key, and local or OpenAI-compatible endpoints. That gives one interface several different cost and privacy paths.

Projects add reusable instructions, context, and separate chat history. Skills and commands handle repeatable work. Quick Ask covers smaller edits that do not need a full agent session.

The strength is coverage. The risk is that a broad tool has more settings and more ways to include the wrong context. Check the selected provider, the agent permission, and the project scope before using it on sensitive notes.

I have a separate breakdown of what Obsidian Copilot does if this is the path you are considering.

Smart Connections is for retrieval first

Smart Connections finds notes and excerpts that are related by meaning, even when they do not share the same exact words.

Its current core plugin ships with a local embedding model. The plugin listing says the default path needs no API key, works offline, and keeps the note index on the device.

This is the smallest useful AI layer for many vaults. It helps with the moment when you know you wrote something but cannot remember the filename, tag, or phrase you used.

It is not the same as an agent. Smart Connections can surface relevant material while you write, but its core job is not to reorganize folders or run a multi-step project across the vault.

If retrieval is the real problem, start here. Adding write access before search works reliably gives an agent a larger room in which to get lost.

Claudian puts coding agents inside the vault

Claudian embeds local agent tools such as Claude Code, Codex, OpenCode, Grok, and Pi in an Obsidian sidebar.

The vault becomes the agent’s working directory. The plugin can expose file reading, writing, search, command execution, inline edits, slash commands, and reusable skills.

This is powerful when the vault is an operating system rather than a passive notebook. An agent can update a project plan, trace related records, or apply a controlled edit without a copy-and-paste loop.

The same power raises the review burden. A command-capable desktop agent can affect more than one note. Use version control or a tested backup, keep approval prompts on, and avoid placing secrets in files the agent can read.

Claudian sends the context you attach and its tool outputs to the provider selected through the connected agent. Local storage does not mean local inference unless the selected provider is also local.

AI Agent is a lighter action layer

AI Agent can read, write, and search notes with a smaller interface than the coding-agent plugins.

Its plugin listing supports Gemini, Anthropic, OpenAI, and a local Ollama model. That makes it useful for someone who wants basic folder and note actions without adopting a full coding-agent workflow.

The key choice is still the provider. Ollama can keep inference local when it is configured correctly. A cloud model receives the prompt and any note content included in the request under that provider’s terms.

Lightweight does not mean harmless. Test write actions on a copy of the vault. A simple request to tidy filenames can create a large cleanup job if the agent guesses a taxonomy you never approved.

AI Assistant keeps the scope closer to chat

AI Assistant focuses on direct chat and prompt-based edits using OpenAI and Anthropic models.

It also lists image generation and speech-to-text features through OpenAI services. Those are useful if you want several AI actions in one panel and do not need a coding agent to operate across the vault.

This narrower scope can be a benefit. Asking a model to rewrite selected text is easier to review than asking it to manage a project folder across several turns.

It is still a cloud-provider workflow. Only attach material you are prepared to send to the configured service, and keep API keys in the plugin setting rather than inside normal notes.

Does Obsidian have built-in AI?

Obsidian does not currently include a general AI assistant as a core plugin. Its official core list covers features such as Search, Bases, Canvas, Backlinks, and Properties.

AI features come from community plugins or external tools. That separation is useful because it lets you choose whether the model is local, cloud-based, chat-only, or able to act on files.

It also means Obsidian’s privacy promise does not automatically describe every plugin. Read the selected plugin’s documentation and the model provider’s terms as separate layers.

How to choose without installing all five

Start with the smallest access level that solves the problem.

• Choose Smart Connections when you mainly need semantic retrieval.
• Choose AI Assistant when you mainly want chat and selected-text edits.
• Choose AI Agent when you want basic note and folder actions with several provider choices.
• Choose Copilot when you want one polished workspace for search, chat, projects, and agents.
• Choose Claudian when you already work with a local coding agent and want that tool inside Obsidian.

Do not install overlapping plugins only to see which sidebar looks better. Each extra index, provider key, and write path adds another place to debug privacy, cost, and conflicting commands.

Privacy depends on the whole route

A local vault is only the first part of the data path. The plugin may build a local index, send selected text to a cloud model, call an existing agent login, or use a hosted backend.

Before choosing, write down four things.

1. Which files can the plugin read?
2. Which files can it change?
3. Which provider receives prompts and context?
4. Can you review and reverse an action?

Keep a backup before the first write test. Start with one disposable folder. Confirm that excluded notes stay excluded, then inspect the actual diff after each agent action.

If you need narrower external access, an Obsidian MCP setup can expose specific tools to an AI client. It takes more setup, but the boundary can be easier to reason about than a broad in-vault agent.

My practical starting point

I would start with retrieval. A vault that cannot surface the right note will not become reliable because an agent can edit it faster.

Use Smart Connections or Obsidian’s own search to prove the structure works. Add a chat or editing plugin only when you can name the repeated task it will remove.

Move to Copilot or Claudian when a real multi-step workflow justifies write and command access. Keep human approval at the point where the agent changes public, financial, or hard-to-recover records.

That is also how I treat AI in a second brain. The model should improve retrieval and execution without becoming the only place that remembers why a decision was made.

Common questions before adding AI to Obsidian

What is the best free Obsidian AI plugin? Smart Connections is a strong starting point for local semantic search. Copilot can also work without a Copilot license when you bring a supported provider, agent account, or local model.

Can an Obsidian AI plugin work offline? Smart Connections uses a local embedding model by default. Other plugins can use local models such as Ollama, but you must confirm that every selected feature follows the local path.

Can Codex work inside Obsidian? Current Copilot and Claudian listings both include Codex paths on desktop. They use different integrations and permission models, so review the setup before pointing either one at an important vault.

Will an AI plugin train on my notes? The answer depends on the provider and route you choose. A plugin may keep its index local while still sending selected context to a cloud model. Check both policies.

Do I need AI in Obsidian? No. If normal search, links, and a clear folder structure already retrieve what you need, an AI plugin may add cost and risk without removing a real bottleneck.

Build the boundary before the automation

The useful question is not which plugin has the longest feature list. It is what the plugin may read, what it may change, and how you will verify the result.

My Tandri setup uses the same rule. AI can search, challenge, and prepare changes. Public or irreversible actions still stop at a visible approval gate.

Choose one job, one plugin, and one test folder. If that small loop works, expand it. If it does not, you have one setting to remove instead of five assistants arguing inside the same vault.