Obsidian AI Setup: Connect a Model and Choose Its Note Access

Obsidian does not include a built-in AI assistant. AI comes through community plugins — extensions the community builds and you install through Obsidian's built-in plugin browser. Each plugin connects to a different model or service and gives the assistant a different level of access to your vault.

Choosing a plugin means choosing a model provider (OpenAI, Anthropic, Google, or a local server like Ollama), deciding what the assistant can read and edit, and testing that access on notes you are comfortable with. Tools like fenbs keep the working instructions and project context that assistants need across sessions, outside any single note app.

For how an AI brain in Obsidian fits into a broader knowledge setup, see AI brain Obsidian and Obsidian AI plugin.

Choose a plugin for your use case

Several community plugins bring AI into Obsidian. The right one depends on whether you want a chat panel, inline text generation, or semantic search across your notes.

• Copilot for Obsidian: a chat panel inside Obsidian that can read notes you paste or include. Connects to OpenAI, Anthropic, Google, and local Ollama models. Good for asking questions about specific notes.
• Smart Connections: builds a local vector index of your vault and surfaces related notes as you write. Supports OpenAI embeddings and local models for the embedding step. Good for finding connections between ideas.
• Text Generator: generates text inline from a cursor position using a template system. Connects to OpenAI and compatible endpoints. Good for drafting and expanding notes in place.

Community plugins are not reviewed by Obsidian before publication. Before installing any plugin, check its GitHub repository: look at the source code or at least the README, review recent issues, and check when it was last updated.

Install a plugin and open its settings

1. Open Obsidian Settings, go to Community plugins, and turn off Restricted mode if you have not already.
2. Click Browse, search for the plugin by name, and click Install, then Enable.
3. Open the plugin's settings panel. You will see fields for an API key or a model server address, a model selection, and usually a vault access setting.
4. Do not close the settings yet — the next step is configuring the connection.

Configure an API key or a local server

For cloud models, you need an API key from the provider. Get one from the provider's developer console: OpenAI at platform.openai.com, Anthropic at console.anthropic.com, Google AI Studio for Gemini. Paste the key into the plugin's settings. The key is stored in your Obsidian configuration folder on disk; treat it like a password.

For local models using Ollama, set the model endpoint to http://localhost:11434 and choose a model you have already pulled with Ollama. The plugin connects to the local server instead of a cloud API, and no data leaves your machine. See Obsidian AI pricing for a cost comparison between routes.

Some plugins let you switch providers per conversation. Start with the connection that is simplest to test, then add others once the basic setup works.

Set which notes the plugin can see

By default, most chat-style plugins only see notes you explicitly include in the conversation — either by pasting content, using an include command, or opening a note before the chat session. They do not automatically read your entire vault.

Semantic search plugins like Smart Connections work differently: they index all notes they have access to. Check the plugin's settings for a path filter or folder exclusion option. Use it to exclude any folder that contains sensitive information, personal data, or notes you do not want indexed.

For questions about data handling and what leaves your machine, see the plugin's documentation and the privacy policy of the model provider you connect. For Obsidian AI security considerations, see Obsidian AI security.

Test the connection before relying on it

Start with a note you do not mind sharing with the model provider. Ask the assistant to summarize it, then ask a question about a specific detail it contains. If the summary is wrong or it does not find the detail, check whether the note was actually included in the context — some plugins require an explicit include step.

Test any vault-reading feature on a small, unimportant folder first. If the assistant returns results about the wrong notes, review the folder path settings. If you see unexpectedly slow responses, check whether the model is loading from a local server or making a cloud API call.

How to do this with fenbs

1. Create a free account and add the project you are working on to a kanban board.
2. Add the project's rules and decisions to Decisions and rules in fenbs: the things every assistant needs to know before it starts working.
3. Connect a supported MCP client (Cursor, Claude, Codex CLI) to fenbs from /integrations to keep the project record separate from your Obsidian vault.
4. Use Obsidian and its AI plugin for writing and note management. Use fenbs for the work state: tasks, decisions, rules, and where the project stands.
5. Before you finish a session, save progress in fenbs. The next session starts from the current state, whichever tool picks it up.

For AI use at work across multiple tools, see how to use AI at work.