SystemSculpt
Install free (opens Obsidian's plugin directory in a new tab)

Ollama in Obsidian: Plugin Settings, Models and Fixes

Published 7 min read

On this page

To use Ollama in Obsidian, install Ollama on your computer, pull a model, and add Ollama as a provider in a community plugin such as Copilot or Local GPT. Ollama serves the model at http://localhost:11434, and the plugin sends your prompts there instead of to a cloud AI.

Plugins with a documented Ollama setup:

  • Copilot has a built-in Ollama provider for Quick Chat, with no license needed.
  • Local GPT runs actions on selected text. It reads Ollama through the AI Providers plugin.
  • Smart Composer lists Ollama for chat and for vault search embeddings. Its README says it is no longer under active development.

We make SystemSculpt, an Obsidian AI plugin that doesn't connect to Ollama, so this guide recommends other plugins. Settings below come from each plugin's docs, checked November 3, 2026.

Which setup fits

You wantPluginOllama models to pull
Chat in a side panelCopilotOne chat model
Summarize or fix selected textLocal GPT with AI ProvidersOne chat model
Answers that use linked notes and PDFsLocal GPT with AI ProvidersA chat model and nomic-embed-text
Chat plus search across the vaultSmart ComposerA chat model and an embedding model

Step 1: Install Ollama and pull models

  1. Download Ollama from ollama.com/download and start it.
  2. Open a terminal and run curl http://localhost:11434. The reply Ollama is running means plugins can reach it.
  3. Pull a chat model, for example ollama pull llama3.2.
  4. Run ollama list and copy the exact name, tag included. Plugins match on that name.

Which model to pull

Sizes come from each model's page in Ollama's library.

JobPullDownload size
Laptop with little memoryollama pull llama3.2:1b1.3 GB
Everyday chat and rewritingollama pull llama3.2 (the 3B version)2.0 GB
Chat that can read imagesollama pull gemma3 (the 4B version)3.4 GB
Stronger answers, tool useollama pull qwen3 (the 8B version)5.2 GB
Reasoning on a well-equipped machineollama pull gpt-oss:20b14 GB
Embeddings for vault search, Englishollama pull nomic-embed-text274 MB
Embeddings, other languagesollama pull bge-m31.2 GB

An embedding model powers search and can't chat, so pull a chat model too. If a plugin's agent or tools matter to you, pick a model tagged tools in the library, such as qwen3, llama3.2 or gpt-oss.

Step 2: Add Ollama to a plugin

Copilot

Copilot's provider guide gives these steps:

  1. Open Settings, Copilot, BYOK.
  2. Select Add a provider and choose Ollama from the Self-host group. This does not need a Self-Host license.
  3. Copilot fills in http://localhost:11434/v1. Leave the API key empty.
  4. Pick your model from the list, or type its name. Select Test, then Save.

New models are turned on for Quick Chat and for Copilot's OpenCode agent by default. Change that under Basic, Agents, Quick Chat. If Test works but Quick Chat can't send, edit the provider and turn on Enable CORS. Replies then arrive in one piece instead of streaming.

Copilot's vault search now runs through Miyo, a separate app. With a small local model, Copilot suggests lowering Maximum results in the Miyo tab.

Local GPT and AI Providers

Install both plugins. In AI Providers:

  1. Add a provider and set Provider type to Ollama.
  2. Click the refresh button and choose your model.

In Local GPT, choose that provider. Select text, then open the context menu or the Action Palette and run Summarize text or another action. For answers that draw on links, backlinks and PDFs, pull nomic-embed-text and select it as the embedding provider in Local GPT's settings.

Smart Composer

Per Smart Composer's documentation, Ollama needs no API key:

  1. Open Settings, Smart Composer.
  2. Add the Ollama provider with the base URL http://127.0.0.1:11434.
  3. Add models using their Ollama names.

Its embedding list includes nomic-embed-text, mxbai-embed-large and bge-m3 from Ollama, for vault search.

Fix CORS and 403 errors with OLLAMA_ORIGINS

Ollama only answers web requests from origins it trusts. Current Ollama source already trusts app:// origins, so update Ollama first. If a plugin still gets a 403 or streaming fails, set OLLAMA_ORIGINS to *, as the AI Providers README suggests. Ollama's FAQ explains how to set its variables on each system:

  • Mac: run launchctl setenv OLLAMA_ORIGINS "*", then quit and reopen Ollama.
  • Windows: quit Ollama from the taskbar. Search Settings for environment variables and choose Edit environment variables for your account. Add OLLAMA_ORIGINS with the value *, then start Ollama from the Start menu.
  • Linux: run systemctl edit ollama.service and add Environment="OLLAMA_ORIGINS=*" under [Service]. Then run systemctl daemon-reload and systemctl restart ollama.

* accepts requests from every origin. Remove the variable if you stop needing it.

Troubleshooting

SymptomLikely causeFix
Connection refused or Test failsOllama isn't runningStart the Ollama app, then check curl http://localhost:11434
403 Forbidden, or streaming failsOrigin not allowedUpdate Ollama, set OLLAMA_ORIGINS, or turn on Enable CORS in Copilot
model '...' not foundName doesn't matchCopy the name from ollama list, tag included, or pull the model
First answer is slowModel is loading into memoryWait. Ollama keeps it loaded for 5 minutes by default
Every answer is slowModel is too big or runs on the CPURun ollama ps and check PROCESSOR, or pull a smaller model
Long notes get ignoredSmall context windowRaise context length in the Ollama app settings; it uses more memory
Vault search returns nothingNo embedding model setPull nomic-embed-text and select it as the plugin's embedding model

Ollama's docs say context defaults to 4K tokens on machines with under 24 GiB of VRAM. Agent and tool work needs at least 64,000.

What leaves your computer

A model pulled with ollama pull runs on your machine. Ollama's FAQ says it doesn't see your prompts when you run locally. Three places can still send notes elsewhere:

  • Ollama cloud models. Names ending in -cloud run on Ollama's servers. Set OLLAMA_NO_CLOUD=1 and restart to turn cloud features off.
  • Other providers in the plugin. If you also add a cloud key, each chat goes to whichever model you pick.
  • Hosted plugin features. Copilot's hosted models send requests to its developer, Brevilabs. Stay on your Ollama provider to avoid that.

Ollama and SystemSculpt

SystemSculpt can't use Ollama by itself. It runs your installed Claude Code or Codex on desktop, free and with no SystemSculpt account, or its own hosted AI.

Codex can use Ollama through its own settings. OpenAI's Codex docs describe codex --oss for local providers such as Ollama. They also describe model_provider in ~/.codex/config.toml for setting a default. SystemSculpt's Codex mode follows those settings when you leave the model on Codex default. How well that works depends on the model. Our Codex in Obsidian guide covers the rest.

Questions

Does Obsidian support Ollama? Not by itself. Obsidian has no AI built in. A community plugin such as Copilot, Local GPT or Smart Composer connects it to Ollama.

What is the best Obsidian plugin for Ollama? Copilot, for chat with a provider screen that fills in the Ollama address for you. Local GPT is best for quick actions on selected text. Smart Composer adds vault search but is no longer actively developed.

Which Ollama model should I use with Copilot? Start with llama3.2 on most laptops, or llama3.2:1b if memory is tight. Move to qwen3 if your machine keeps up and you want tool use.

Does Smart Connections work with Ollama? It doesn't need Ollama. Its README says it ships a built-in local embedding model with no extra app. Model connections, cloud and local, moved to Smart Chat Pro, one of its Pro plugins for project supporters.

Why does my plugin get a 403 from Ollama? Ollama rejected the request's origin. Update Ollama, then set OLLAMA_ORIGINS as shown above and restart it.

Can I use Ollama with MCP in Obsidian? Ollama is the model server, not the MCP client. The plugin handles MCP. Smart Composer connects to MCP servers and lists Ollama as a provider. Choose a model tagged tools. See our Obsidian MCP server guide.

Is Ollama free? Running models on your own computer is free. Its cloud models are a separate service; check the current price.

Does it work on my phone? Ollama runs on a computer, not on iPhone or Android. Copilot's Quick Chat works on mobile, but it needs an Ollama server your phone can reach.

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.

TopicsObsidianObsidian AI

Try SystemSculpt inside your vault

SystemSculpt adds first-party AI chat, semantic search, transcription, and document workflows to Obsidian.

View pricing