Cortex Local is a small Obsidian side panel for working with notes using a local AI model.
The main idea is simple:
- open a note
- ask Cortex Local about it
- use one-click actions like Summarize or Improve writing
- approve edits before Cortex changes the file
It is Ollama-first, so you can use it without an API key.
Status: early local plugin. It is not in the Obsidian community plugin store yet. Name note: there is already a community plugin called Cortex Chat, so this project uses Cortex Local and the plugin id
cortex-local.
- Chat about the note you have open.
- Improve or rewrite the active note after asking you first.
- Use saved prompt buttons for common tasks.
- Import a ChatGPT
conversations.jsonexport into Markdown notes. - Optionally connect ACP-compatible agents like Claude Code or Codex.
Download or build these three files:
main.jsmanifest.jsonstyles.css
Create this folder inside your vault:
<your-vault>/.obsidian/plugins/cortex-local/
Put the three files there, then open Obsidian:
- Settings
- Community plugins
- Turn off Restricted mode if needed
- Enable Cortex Local
- Reload Obsidian if the brain icon does not appear
Cortex Local works best with Ollama running locally.
Install Ollama, then run:
ollama pull llama3.2
ollama serveCheck your model:
ollama listDefault Cortex Local settings:
Address: http://127.0.0.1:11434
Model: llama3.2:latest
Click the brain icon in the left ribbon, or use the command palette:
Cortex Local: Open Cortex Local
Try:
- "Summarize this note"
- "Explain this simply"
- "Improve the active note's clarity"
For note edits, Cortex Local shows a permission card before replacing the file. Pick Apply edit or Keep original.
Local Ollama requests stay on your machine.
If you add an external ACP agent, that agent may use its own CLI, account, or API. Be careful with private notes. Cortex keeps open-note context off for external agents unless you enable it.
File access is clamped to the Obsidian vault. Writes ask before changing a file unless you turn on auto-save behavior in settings.
Do not put API keys in Cortex settings. Use your operating system environment instead.
Cortex Local can import ChatGPT exports:
- Export your ChatGPT data from ChatGPT settings.
- Unzip the export.
- In Obsidian, run:
Cortex Local: Import ChatGPT export
Pick conversations.json.
By default, notes go into:
ChatGPT/
Each conversation becomes one Markdown note.
Related standalone tool:
npm install
npm run buildThe build writes main.js in the project root.
For manual install, copy:
main.js
manifest.json
styles.css
into:
<your-vault>/.obsidian/plugins/cortex-local/
If Cortex Local says Ollama is not available:
ollama serve
ollama listIf the plugin does not show in Obsidian:
- check the folder name is exactly
cortex-local - check
manifest.jsonis in that folder - reload Obsidian
If note edits feel risky, leave write approval on. Cortex Local will ask before replacing the current note.