Local LLM in Obsidian: Ollama and LM Studio, Step by Step

To use a local LLM in Obsidian, you need two things: a program that runs the model on your computer, and a plugin that connects Obsidian to it. Obsidian has no AI of its own. The two common model programs are Ollama and LM Studio, both free. Your notes and prompts then stay on your machine.

The plugins worth starting with, all free:

• Copilot for chat with your vault and an agent, with Ollama or LM Studio as a provider.
• Local GPT for quick actions on selected text, such as summarize or fix grammar.
• Vault Operator for an agent that plans, reads and edits notes, asking before each change.
• Smart Connections for related notes. It ships its own small local model, so it needs no Ollama at all.

We make SystemSculpt, an Obsidian AI plugin that doesn't run local models, so this guide recommends other plugins. Plugin facts checked November 3, 2026, against each plugin's README and the Obsidian plugin directory.

Which setup fits

Step 1: Run a model on your computer

Pick one program. Both serve the model at an address on your computer that plugins call, the same way they would call a cloud API.

Ollama runs in the background and is driven from a terminal:

1. Download it from ollama.com and open it.
2. In a terminal, download a model from Ollama's library, for example ollama pull gemma3.
3. Run ollama list to see the exact model name. Plugins need that name.

Ollama listens at http://localhost:11434.

LM Studio is an app with a window, so you can do everything without a terminal:

1. Download it from lmstudio.ai and open it.
2. Search for a model in the app and download it.
3. Start its local server in the app's developer view.

LM Studio listens at http://localhost:1234/v1.

Smaller models answer faster and need less memory. Start with a small one, and move up if your computer keeps up.

Step 2: Connect a plugin

Copilot: chat with your vault

Install Copilot from Settings, Community plugins, Browse. Then, per Copilot's provider guide:

1. Open Settings, Copilot, BYOK.
2. Select Add a provider, then choose Ollama or LM Studio. Copilot fills in the local address, and no API key or license is needed.
3. Select or type your model name, select Test, then Save.

Pick the model in Quick Chat, or in Copilot's OpenCode agent. If Test passes but chat can't send a message, edit the provider and turn on Enable CORS. Answers then arrive all at once instead of word by word. With LM Studio, you can turn on CORS in LM Studio instead and keep streaming.

Local GPT: actions on selected text

Local GPT gets its models from a second plugin, AI Providers, by the same developer. Install both, then:

1. In AI Providers, add a provider and set Provider type to Ollama, or to LM Studio with the URL http://localhost:1234/v1.
2. In Local GPT, choose that provider.
3. Select text in a note, open the context menu or the Action Palette, and pick an action such as Summarize text.

If streaming fails with Ollama on a Mac, the AI Providers README suggests running launchctl setenv OLLAMA_ORIGINS "*" in a terminal, then restarting Ollama.

For answers that use your linked notes, Local GPT also needs an embedding model. Its README suggests ollama pull nomic-embed-text for English.

Vault Operator: an agent on a local model

Vault Operator, which was called Obsilo Agent, runs a full agent inside Obsidian. It searches, reads and writes notes, asks before each write by default, and keeps checkpoints so you can undo a change. It works with cloud providers or fully offline with Ollama or LM Studio. Add the provider under Settings, Vault Operator, Providers. It is desktop only.

Small local models are less reliable at multi-step agent work than large cloud models. Try a simple task first, such as "What are my most-linked notes?"

Smart Connections: related notes with no setup

Smart Connections finds notes related to the one you're writing. Its built-in model starts indexing as soon as you enable it, with no API key and no extra app. Chat moved to a separate plugin, Smart Chat, and its local and cloud model connections are in Smart Chat Pro.

What stays on your computer

With a local model, the model runs on your machine, so your notes and prompts don't go to an AI company. Two things to check:

• Optional cloud features. Some plugins also offer hosted models or services. Copilot's hosted models and paid features go to its own servers. Stay on your local provider to keep everything local.
• Every provider you add. A plugin sends your notes to whichever provider you choose. If you add a cloud key next to your local model, check which one each chat uses.

When a local model isn't enough

Local models trade quality and speed for privacy. On a laptop, a small model is fine for summaries and rewrites, and weaker at long, multi-step work across many notes.

If you already pay for Claude or ChatGPT, you can use that plan in Obsidian instead. SystemSculpt runs your installed Claude Code or Codex inside Obsidian for free, with no SystemSculpt account. Our guides cover Claude Code in Obsidian and Codex in Obsidian. For your phone, SystemSculpt's hosted AI gives eligible new accounts 500 one-time starter credits, then costs $9 a month.

Questions

Can Obsidian use a local LLM? Yes, through a community plugin. Run the model with Ollama or LM Studio, then connect a plugin such as Copilot, Local GPT or Vault Operator to it.

What is the best local LLM plugin for Obsidian? Copilot, for chat with your vault and an agent. Local GPT is simpler if you only want actions on selected text. Vault Operator is the agent built for local models. Smart Connections covers related notes with its own built-in model.

Do I need Ollama or LM Studio? For chat and writing, yes; the plugins don't include a model. Smart Connections is the exception for related notes: its small model is built in.

Is it free? Yes. Ollama, LM Studio, Copilot's free plan, Local GPT, AI Providers, Vault Operator and Smart Connections cost nothing. The only cost is your computer's memory and battery.

Does it work offline? Yes, once the model is downloaded. Ollama and LM Studio run without internet, and so do the plugins' local providers.

Does it work on iPhone or Android? Not with a model running on your phone. Ollama and LM Studio run on a computer, so these setups are for Obsidian on desktop.

Can SystemSculpt use a local model? Not by itself. SystemSculpt runs your installed Claude Code or Codex, or its hosted AI. Codex can be set up to use Ollama or LM Studio in its own settings, and SystemSculpt's Codex mode follows those settings when you leave the model on Codex default. How well that works depends on the model.

Install SystemSculpt free from the Obsidian plugin directory to run Claude Code or Codex in your vault, or compare its plans for hosted AI.

Sources

Checked November 3, 2026.

• Copilot: README, LLM providers guide
• Local GPT: README; AI Providers: README
• Vault Operator: README, documentation
• Smart Connections: README
• Ollama, LM Studio
• SystemSculpt: Choose your AI, pricing, privacy policy