Local LLM Helper

Private AI for your Obsidian vault.
Chat with your notes, find related ideas, run review workflows, and transform text with your own local or OpenAI-compatible models.

Privacy and capabilities · Contributing · Releases

Why Use It?

Local LLM Helper turns Obsidian into a private AI workspace without forcing your notes into someone else's app. It works with Ollama, LM Studio, vLLM, LocalAI, OpenAI, and most servers that expose OpenAI-compatible endpoints.

• Keep your notes in Obsidian - ask questions, summarize, rewrite, and review without changing tools.
• Use local models when privacy matters - point the plugin at your own chat and embedding server.
• Search by meaning, not just keywords - index your vault and retrieve the notes that actually matter.
• Move from chat to action - generate review notes, task updates, and project summaries behind explicit approval cards.

Highlights

See It In Action

Chat With Your Notes

Ask questions against your indexed vault, inspect sources, and narrow the context when you need a more precise answer.

Fast Commands From Obsidian

Run text transformations, chats, searches, note indexing, and workflows from the ribbon menu or command palette.

Settings Built For Real Model Setups

Configure providers, chat models, embedding models, personas, reasoning extraction, RAG indexing, saved prompts, workflow defaults, and web search.

Features

🔌 Multi-Provider Support

Works with any server that supports the OpenAI API format (/v1/chat/completions):

• Ollama - Run open-source models locally
• OpenAI - Use GPT models with your API key
• LM Studio - Local inference with a GUI
• vLLM, LocalAI, text-generation-webui, and more

Switch providers anytime from settings. Configure temperature, max tokens, and streaming.

📚 Notes RAG

Index your vault for semantic search and AI-powered Q&A:

• Smart chunking - Documents split with overlap for better context.
• Incremental indexing - Only re-indexes changed files.
• Content preprocessing - Strips frontmatter and cleans markdown.
• Attachment-aware retrieval - Index PDF attachments by default, with opt-in local OCR for images and scanned PDFs.
• Multiple embedding providers - Works with OpenAI-compatible embedding endpoints.
• Configurable retrieval - Tune how many chunks are sent to the model.

🧭 Related Notes Sidebar

Keep semantic context visible while you write:

• Uses selected text when you highlight a passage.
• Falls back to the full active note when nothing is selected.
• Shows related notes with scores, paths, and previews.
• Opens any match instantly or starts scoped RAG chat with selected notes.

🧩 Manual Workflow Automation

Run useful review recipes without giving the model autonomous write access:

• Weekly review - Turn recent or scoped notes into wins, open loops, and next actions.
• Meeting notes to tasks - Extract action items and follow-ups into a target note.
• Project status summary - Summarize progress, risks, blockers, and next steps.
• Approval-first writes - Every generated note change is reviewed before it is applied.

✍️ Text Processing

Transform selected text with AI-powered commands:

All commands are available via Command Palette (Cmd/Ctrl + P) or the ribbon menu.

💬 Chat Interfaces

LLM Chat for general conversation:

• Remembers recent exchanges for context.
• Supports all configured personas.
• Can propose approval-gated Vault Actions when enabled.
• Opens in a docked sidebar so your notes stay visible while you chat.

RAG Chat for note-grounded answers:

• Searches your indexed notes semantically.
• Shows clickable sources that open notes and indexed attachments.
• Supports vault, note, folder, tag, and selected-note scopes.
• Uses the same sidebar as general chat; popup variants remain available from the command palette.

🌐 Web Integration

• Web Search - Search the web using Tavily, Brave API, or a self-hosted SearXNG instance.
• News Search - Get recent news on any topic.
• SearXNG Support - Configure an instance URL in settings; the instance must enable JSON output with json in search.formats.

🎭 Personas

Customize AI behavior with 12 built-in personas — or create your own:

• Software Developer, Physics Expert, Fitness Expert, Stoic Philosopher
• Product Manager, Technical Writer, Creative Writer
• TPM, Engineering Manager, Executive, Office Assistant

Edit any persona's system prompt directly in settings. Create custom personas with your own name and instructions. Restore defaults anytime.

📋 Saved Prompts

Save frequently-used prompts and run them instantly:

• Each saved prompt registers as a command in the palette — assign hotkeys to your favorites
• Fuzzy search picker to quickly find and run any saved prompt
• Full CRUD in settings: create, edit, rename, delete
• Existing single "Custom Prompt" field still works alongside saved prompts

🧠 Reasoning Extraction

Models like DeepSeek and Qwen wrap their thinking process in <think> blocks. Enable reasoning extraction to automatically strip these from output:

• Strips <think>, <reasoning>, and <thought> blocks by default
• Configurable markers via JSON for custom formats
• Applied to text commands, chat, and RAG responses

🔍 Model Picker

Browse available models directly from your server:

• Browse button next to Chat and Embedding model fields
• Fetches from /v1/models endpoint (chat browse uses the chat/default server; embedding browse uses the embedding server override when set)
• Searchable picker modal — no more guessing model names
• Manual text entry always available as fallback

⚙️ Organized Settings

Settings organized into clear sections:

• Connection & Provider
• Models (with Browse)
• Chat & Personas
• Output & Reasoning
• Custom Prompt & Saved Prompts
• Notes Index (RAG)
• Workflow Automation
• Integrations
• About & Changelog

🎯 Command Organization

All commands use clear prefixes for easy discovery:

• Text: - Text transformation commands
• Prompt: - Your saved custom prompts (hotkey-assignable)
• Chat: - Open chat interfaces
• Workflow: - Run review workflows
• Web: - Web and news search
• Notes: - RAG indexing and management
• Settings: - Plugin configuration

Installation

From Community Plugins (Recommended)

1. Open Obsidian Settings → Community Plugins
2. Click "Browse" and search for "Local LLM Helper"
3. Install and enable the plugin

Manual Installation

1. Download the latest release from GitHub Releases
2. Extract to your vault's .obsidian/plugins/ folder
3. Enable in Settings → Community Plugins

Quick Start

1. Configure Your Provider

Go to Settings → Local LLM Helper and choose your provider:

For Ollama:

• Server: http://localhost:11434
• Model: Click Browse to pick from available models, or type llama3.2
• Embedding Model: mxbai-embed-large

For OpenAI:

• Server: https://api.openai.com
• API Key: Your OpenAI API key
• Model: gpt-4 or gpt-3.5-turbo

For LM Studio:

• Server: http://localhost:1234
• Model: Click Browse to see loaded models, or leave blank for default

2. Try Text Commands

1. Select some text in a note
2. Open Command Palette (Cmd/Ctrl + P)
3. Type "Text:" to see available commands
4. Choose an action like "Summarize" or "Make Professional"

3. Index Your Notes

1. Command Palette → "Notes: Index notes for RAG"
2. Wait for indexing to complete
3. Command Palette → "Chat: Notes (RAG)" to chat with your notes in the sidebar
4. Command Palette → "Notes: Open related notes" to keep nearby notes visible while writing

4. Run a Workflow

1. Command Palette → "Workflow: Run workflow..."
2. Choose a built-in recipe
3. Confirm the note scope and output target
4. Review the generated approval card before writing anything to your vault

Notes:

• Workflows use the same indexed-note context as RAG chat, so run Notes: Index notes for RAG first if your vault changed.
• PDF attachments are indexed by default; image OCR and scanned-PDF OCR can be enabled in Settings → Notes Index (RAG).
• Meeting notes to tasks and Project status summary require a target note to append into.
• Workflow approvals still work even if the chat-only Vault Actions toggle is off.

Changelog

v2.4.5

Dockable Chat Sidebar

• Moved general and notes chat into a right-sidebar workspace view so notes stay visible while chatting.
• Added a multiline composer with target context, stop state, inline readiness/error banners, and a verification reminder.
• Improved source citations with note titles and paths that open in the main editor.

v2.4.2

Plugin Review Cleanup

• Published a fresh release with matching manifest metadata.
• Added build provenance attestations for release assets.
• Clarified optional auto-index behavior and kept it disabled by default.

Workflow Automation

• Added a manual workflow runner for weekly review, meeting notes to tasks, and project status summary recipes
• Reused note, folder, and tag scopes across RAG chat and workflow runs
• Kept all workflow note writes behind explicit approval cards

Related Notes Sidebar

• Added a persistent "Related Notes" workspace view
• The sidebar updates from the current note or selected text
• Open matching notes directly or start RAG chat with selected related notes

v2.4.0

Saved Prompts

• Save frequently-used prompts with title and text
• Each saved prompt registers as a command — assign hotkeys to your favorites
• Fuzzy-searchable picker modal (Text: Run saved prompt...)
• Full CRUD in settings: create, edit, rename, delete

Persona Editing

• View and edit any persona's system prompt in settings
• Create fully custom personas with your own name and instructions
• Delete custom personas, restore all defaults with one click
• Same backward-compatible persona keys — existing settings just work

Reasoning Extraction

• Toggle to strip <think>, <reasoning>, <thought> blocks from LLM output
• Useful for DeepSeek, Qwen, and other models that expose chain-of-thought
• Configurable markers via JSON for custom formats
• Applied to text commands, general chat, and RAG chat

Model Picker

• Browse button next to Chat and Embedding model fields
• Fetches chat models from the chat/default server and embedding models from the embedding server override when set
• Searchable picker modal — works with Ollama, LM Studio, vLLM, and others
• Manual text entry always available as fallback

Bug Fixes

• Fixed server URL normalization to prevent missing protocol errors
• Resolved security vulnerabilities and cleaned up dependencies

v2.3.1

New Features

• Redesigned RAG Chat: New interface with welcome message, example queries, and clickable sources
• Changelog in Settings: View version history anytime from Settings → About

RAG Improvements

• Smarter chunking with overlap for better context preservation
• Incremental indexing - only re-indexes changed files
• Content preprocessing - strips frontmatter and cleans markdown
• Better error messages when notes aren't indexed

UI/UX

• Commands organized with prefixes (Text:, Chat:, Web:, Notes:)
• Ribbon menu grouped logically with separators
• Settings page organized into 7 clear sections
• All prompts improved for better LLM output
• Persona prompts rewritten to be more actionable

v2.3.0

• Edit with Prompt: Edit selected text with preset or custom prompts
• Security Updates: Fixed dependency vulnerabilities
• Better Error Messages: Clearer embedding failure messages

v2.2.1

• Fixed re-embedding issue on every restart
• Proper persistent storage for embeddings
• Storage diagnostics command

v2.2.0

• Multi-provider support (Ollama, OpenAI, LM Studio)
• Easy provider switching in settings
• Configurable temperature and max tokens

Requirements

• Obsidian 1.7.0 or later
• Any LLM server with OpenAI-compatible API (Ollama, LM Studio, OpenAI, vLLM, etc.)

Support

• Report Issues
• Buy Me a Coffee

License

MIT License - see LICENSE for details.