LLM Text Assistant

🤖 LLM Text Assistant – Obsidian Plugin

Process, translate and refine text in your Obsidian notes with LLMs, directly in the editor

This is the Obsidian edition of LLM Text Assistant — the browser extension (Chrome/Firefox/Thunderbird) by the same author, with the same actions, prompts and streaming behaviour, adapted to the Obsidian editor.

✨ Key Features

📌 Core Features

• Four built-in actions: 🌐 Translate, ✍️ Expand, 📋 Summarize, ✅ Grammar & Spelling, available via the editor context menu (right-click), the command palette, the ribbon icon ✨ and its action menu
• Custom actions: Define your own actions with emoji, title and system prompt
• Free Prompt (Chat mode): Open a chat window with the current selection as context and give iterative instructions with full context; apply the result back into the note — with ready-made preset chips (Shorter / More formal / As email / Bullet list) and an optional note context (note title, vault path, surrounding text)
• Copy responses: Hover any response for its ⧉ copy button, use "Copy last response", or press Ctrl/Cmd+C with nothing selected
• Streaming results (live typing): Text appears token by token as the model generates it, no waiting for the full response
• Selection or whole note: Processes the selection — or the entire note when nothing is selected
• Review before replacing (optional): Show an old→new diff and only write the result after you confirm

⌨️ UX

• Undo toast: After every replacement a toast appears with an Undo button (8 seconds)
• Cancel requests: Abort a running request by clicking the "Processing…" toast or the stop button in chat, already received text is kept
• Native Obsidian hotkeys: Every action — the four built-in ones, every custom action and the free prompt — is a command; assign hotkeys via Obsidian's Hotkeys settings
• Robust error handling: Configurable timeout (default 60s), automatic retries on network errors and rate limits (429/5xx), clear error messages (e.g. "check your API key" on 401)

🔌 Provider Support

• Works with any OpenAI-compatible Chat Completions API:
  • Cloud: OpenAI, Mistral, Groq, Google Gemini (OpenAI-compat endpoint) and more
  • Local: Ollama, LM Studio, llama.cpp, vLLM; no API key required, data never leaves your machine
• Flexible configuration: API URL, API key, model, temperature, timeout
• Model list from the endpoint: "Load models" queries GET <base>/models and offers the available models as autocomplete for the model field
• Customizable system prompts for every built-in action (e.g. {TARGET_LANGUAGE} placeholder for translation)

🌍 Internationalization

• Full UI and default prompts in English and German (auto-selected by Obsidian language)
• Target language for translation is freely configurable

🚀 Quick Start

Prerequisites

• Obsidian 1.8.7 or newer (Restricted Mode off for community plugins)
• An OpenAI-compatible API endpoint (cloud or local)

🛠️ Installation

Install the plugin directly from the Obsidian Community directory:

👉 https://community.obsidian.md/plugins/llm-text-assistant

Or from within Obsidian: Settings → Community Plugins → Browse → search for LLM Text Assistant → Install.

Then:

1. Enable LLM Text Assistant under Settings → Community Plugins
2. Configure your endpoint under Settings → LLM Text Assistant

1. Download main.js, styles.css and manifest.json from the latest release
2. Create the folder <your-vault>/.obsidian/plugins/llm-text-assistant/ and copy the three files into it — or use Obsidian 1.8+: Settings → Community Plugins → Install from files
3. Enable the plugin and configure your endpoint as above

⚙️ Configuration

Open Settings → LLM Text Assistant:

Hotkeys: every action — built-in, custom and the free prompt — appears in Obsidian's command palette; assign hotkeys under Settings → Hotkeys.

Local endpoint examples

📖 Usage

1. Select text in the editor (or select nothing to process the whole note)
2. Choose an action via right-click context menu, command palette (Ctrl/Cmd+P), the ribbon icon ✨ or its action menu
3. The processed text replaces the selection live (streaming)
4. Cancel: click the "Processing…" toast or ⏹ Stop in the chat
5. Undo: use the button in the toast (8 seconds)
6. Copy: in the free prompt chat, hover a response and click ⧉, use "Copy last response", or press Ctrl/Cmd+C with nothing selected

💬 Free Prompt (Chat mode)

Open it via the context menu or the action menu. Your current selection is loaded as context — give instructions one after another, e.g. "Improve the text" or "Translate into English". ✓ Apply writes the result into the editor, ⧉ copies it to the clipboard, ↺ Reset restarts the chat with fresh context.

• Preset chips (Shorter / More formal / As email / Bullet list) drop a ready-made instruction into the input field — they never send on their own
• Note context: The "Note context" switch in the chat header adds the note title, its vault path and the surrounding text to every request. It starts from the setting Pass note context to the chat and can be toggled per chat session

✅ Review before replacing

With Confirm result before replacing enabled, every action shows a Review changes window after the result has streamed in: removed words are struck through in red, added words are highlighted in green. Apply keeps the result (the undo toast still works), Discard (or Esc) restores the original text — nothing is ever written without your confirmation.

🔒 Privacy & Security Notes

• No data collection: The plugin collects nothing. API calls go directly from Obsidian to the endpoint you configured, there is no intermediate server
• API key storage: Your API key is stored in the plugin settings of your vault (.obsidian/plugins/llm-text-assistant/data.json) and only sent as Authorization header to the configured endpoint

🏗️ Development

Prerequisites: Node.js ≥ 20 and npm.

npm install
npm run build      # type-check (tsc) + bundle main.js
npm run typecheck  # TypeScript check only
npm run smoke      # core tests without an Obsidian runtime (streaming, retry, abort, i18n)
npm run dev        # watch mode for development

The smoke tests mock Obsidian's requestUrl via a Node loader hook (scripts/smoke-hook.mjs) — no Obsidian installation required. See AGENTS.md for conventions and release details.

🤝 Contributing

We welcome contributions from the community!

For code changes by third parties, please coordinate with us via email at [email protected] before making any changes.

You can also:

• Open an Issue for bug reports or feature requests
• Submit a Pull Request for improvements
• Help improve documentation

💖 Support the Project

If you find this project useful and would like to support its continued development, you can buy me a coffee! Your support helps me dedicate more time and resources to improving the application and adding new features. While financial support is not required, it is greatly appreciated and helps ensure the project's ongoing maintenance and development.

---

📄 License: GNU GENERAL PUBLIC LICENSE Version 3 (see LICENSE file)