Obsidian Autopilot Backend

A Python tool for processing and enriching your Obsidian notes, providing automatic summarization, task extraction, and semantic knowledge base features.

Recent Updates

Automatic Capability Detection (Latest)

• Ollama models now automatically detect tool calling and reasoning capabilities via programmatic inspection
• No more hardcoded model lists - capabilities are detected by examining model templates through Ollama's API
• Reasoning mode support automatically detected for models like Qwen3, DeepSeek-R1, QwQ, GPT-OSS, and others
• Tool calling support automatically detected for compatible models
• Falls back to known model lists if API detection fails

Key Features

• Note Processing

  • Generate daily and weekly summaries
  • Extract and structure meeting notes
  • Identify action items and add them to Apple Reminders
  • Extract learnings with auto-generated tags

• Knowledge Base

  • Build semantic connections between notes
  • Support for OpenAI or local Ollama embeddings
  • Auto-generate wiki-style links
  • Create backlinks automatically
  • Visualize note connections in graph format

• Flexible AI Backend

  • OpenAI API support for text generation and embeddings
  • Ollama support for local AI processing (both text generation and embeddings)
  • Configurable base URLs for OpenAI-compatible APIs

Meeting Transcripts from Clipboard

You can use the clipboard to process meeting transcripts directly:

1. Copy the full meeting transcript to your clipboard
2. Run the command:

   python main.py notes --from-clipboard

This will:

• Take the transcript from your clipboard
• Process it with the meeting notes prompt
• Intelligently infer a descriptive topic name from the content
• Save a formatted summary to your configured meeting notes directory with filename YYYY-MM-DD_inferred_topic.md

You can also specify a custom prompt file:

python main.py notes --from-clipboard --prompt-file /path/to/custom_prompt.md

Installation

Prerequisites

• Python 3.x
• OpenAI API key (if using OpenAI embeddings)
• Apple Reminders (optional, for task integration)

Quick Start

1. Clone the repository:

   git clone https://github.com/ivo-toby/gpt-notes-to-tasks.git
   cd gpt-notes-to-tasks

2. Create and activate virtual environment:

   python -m venv .venv
   source .venv/bin/activate  # On Unix/macOS
   # or
   .venv\Scripts\activate  # On Windows

3. Install dependencies:

   pip install -r requirements.txt

4. Set up configuration:

   cp config.template.yaml config.yaml
   # Edit config.yaml with your settings

Ollama Setup (Local AI Processing)

For complete local AI processing without external API calls:

1. Install Ollama: Follow instructions at ollama.ai

2. Pull required models:

   # For text generation (summaries, meeting notes, etc.)
   ollama pull llama3.2
   
   # For embeddings (knowledge base)
   ollama pull mxbai-embed-large

3. Configure for Ollama:

   # LLM Inference Configuration
   inference:
     provider: "ollama"
   
     ollama:
       base_url: "http://localhost:11434"
       model: "llama3.2"  # or qwen3, deepseek-r1, mistral, etc.
       temperature: 0.7
       num_ctx: 8192
       num_thread: 4
       timeout: 120
   
       # Reasoning mode (automatically detected for compatible models)
       reasoning:
         enabled: false        # Enable for reasoning-capable models
         save_thinking: false  # Include thinking process in outputs
         log_thinking: false   # Log thinking content for debugging
   
   # Embedding settings
   embeddings:
     model_type: "ollama"
     model_name: "mxbai-embed-large"
     ollama_config:
       base_url: "http://localhost:11434"

4. Start Ollama service:

   ollama serve

This setup provides complete privacy and offline functionality for all AI operations.

Reasoning Mode (Advanced)

Some Ollama models support reasoning mode, which enables chain-of-thought processing:

1. Supported models (automatically detected):

   • Qwen3, Qwen2.5
   • DeepSeek-R1, DeepSeek-V3
   • QwQ, Smallthinker
   • GPT-OSS, Magistral
   • And any other model with reasoning template markers

2. Enable reasoning mode:

   inference:
     ollama:
       reasoning:
         enabled: true         # Enable reasoning mode
         save_thinking: true   # Include thinking process in saved outputs
         log_thinking: false   # Enable for debugging thinking content

3. How it works:

   • Models with reasoning support automatically use extended thinking
   • Thinking tokens can be saved or suppressed based on configuration
   • Tool calling works seamlessly with reasoning mode
   • Capabilities are auto-detected via Ollama's API

CLI Reference

Daily Notes Commands

# Process today's notes
python main.py notes

# Process notes for a specific date
python main.py notes --date 2024-02-16

# Process yesterday's notes
python main.py notes --date yesterday

# Generate weekly summary
python main.py notes --weekly

# Process meeting notes only
python main.py notes --meetingnotes

# Process learnings only
python main.py notes --process-learnings

Options:

• --date: Specify date (YYYY-MM-DD, "today", or "yesterday")
• --dry-run: Preview changes without making modifications
• --skip-reminders: Don't create Apple Reminders tasks
• --replace-summary: Replace existing summaries instead of updating

Knowledge Base Commands

# Initialize knowledge base
python main.py kb --reindex

# Update with recent changes
python main.py kb --update

# Search knowledge base
python main.py kb --query "your search query"

# Search by tag
python main.py kb --find-by-tag "tag-name"

# Search by date
python main.py kb --find-by-date "2024-02-16"

# Analyze single note relationships
python main.py kb --analyze-links "path/to/note.md"

# Show note connections
python main.py kb --show-connections "path/to/note.md"

# Analyze and auto-link all notes
python main.py kb --analyze-all

# Analyze only recently modified notes
python main.py kb --analyze-updated

# View note's semantic structure
python main.py kb --note-structure "path/to/note.md"

Options:

• --limit: Maximum number of results (default: 5)
• --dry-run: Preview changes without modifications
• --note-type: Filter by note type (daily, weekly, meeting, learning, note)
• --graph: Display connections in graph format (if configured)
• --auto-link: Automatically add suggested links without prompting

Common Commands & Howto

Daily Note Taking Workflow

1. Start your day by processing yesterday's notes:

   python main.py notes --date yesterday

2. Throughout the day, take notes in your daily notes file

3. End of day processing:

   # Process today's notes and create summaries
   python main.py notes
   
   # Update knowledge base with new content
   python main.py kb --update
   
   # Analyze new connections
   python main.py kb --analyze-updated --auto-link

Weekly Review Process

1. Generate weekly summary:

   python main.py notes --weekly

2. Process learnings:

   python main.py notes --process-learnings

3. Update knowledge connections:

   python main.py kb --analyze-all --auto-link

Knowledge Base Maintenance

1. Regular updates (daily/after changes):

   python main.py kb --update
   python main.py kb --analyze-updated

2. Full reindex (monthly or after major changes):

   rm -rf ~/Documents/notes/.vector_store
   python main.py kb --reindex
   python main.py kb --analyze-all --auto-link

3. Finding Related Content:

   # Search by content
   python main.py kb --query "project planning" --limit 10
   
   # Find notes with specific tag
   python main.py kb --find-by-tag "project" --note-type meeting
   
   # Show connections for a note
   python main.py kb --show-connections "path/to/note.md" --graph

Troubleshooting Command

Analyze note structure to understand how it's being processed:

python main.py kb --note-structure "path/to/note.md"

Configuration

Copy config.template.yaml to config.yaml and configure your settings:

# Essential paths
notes_base_dir: "~/Documents/notes"
daily_notes_file: "~/Documents/notes/daily/daily.md"
daily_output_dir: "~/Documents/notes/daily"
weekly_output_dir: "~/Documents/notes/weekly"
meeting_notes_output_dir: "~/Documents/notes/meetingnotes"
learnings_file: "~/Documents/notes/learnings/learnings.md"
learnings_output_dir: "~/Documents/notes/learnings"

# LLM Inference Configuration
inference:
  provider: "openai"  # Options: "openai" | "ollama"

  # OpenAI-specific settings
  openai:
    api_key: "your-api-key"
    model: "gpt-4o"
    base_url: "https://api.openai.com/v1"  # Optional: for OpenAI-compatible APIs

  # Ollama-specific settings
  ollama:
    base_url: "http://localhost:11434"
    model: "llama3.2"  # or qwen3, deepseek-r1, mistral, etc.
    temperature: 0.7
    num_ctx: 8192
    num_thread: 4
    timeout: 120

    # Reasoning mode (auto-detected for compatible models)
    reasoning:
      enabled: false        # Enable for reasoning-capable models
      save_thinking: false  # Include thinking process in outputs
      log_thinking: false   # Log thinking content for debugging

# Embedding configuration (recommended local setup)
embeddings:
  model_type: "ollama"
  model_name: "mxbai-embed-large"
  ollama_config:
    base_url: "http://localhost:11434"
    num_ctx: 512
    num_thread: 4

Vector Store Settings

vector_store:
  path: "~/Documents/notes/.vector_store"
  similarity_threshold: 0.60 # For normalized embeddings
  # HNSW index settings - adjust if you get "ef or M is too small" errors
  hnsw_config:
    ef_construction: 800 # Higher = better index quality, slower build (default: 400)
    ef_search: 400 # Higher = more accurate search, slower queries (default: 200)
    m: 256 # Higher = better accuracy, more memory usage (default: 128)

If you're processing a large number of notes or getting HNSW errors, try increasing these values.

Search Thresholds

search:
  thresholds:
    default: 0.60
    tag_search: 0.50
    date_search: 0.50

Usage Guide

Setting Up Your Notes Repository

1. Create basic structure:

   mkdir -p ~/Documents/notes/{daily,weekly,meetingnotes,learnings}

2. Configure .gitignore:

   .vector_store/
   .obsidian/
   .trash/
   *.excalidraw.md
   .DS_Store
   

Daily Workflow

1. Write notes throughout the day
2. Process daily notes:

   python main.py notes

3. Update knowledge base:

   python main.py kb --update

Weekly Workflow

1. Generate weekly summary:

   python main.py notes --weekly

2. Review and process learnings:

   python main.py notes --process-learnings

Technical Details

Embedding Model Comparison

Two primary options are available:

1. OpenAI (text-embedding-3-small)

   • Normalized embeddings (0-1 range)
   • Typical scores: 0.35-0.45
   • Requires API key and internet connection

2. Ollama (mxbai-embed-large)

   • Local processing
   • Better handling of technical content
   • Typical scores: 0.60-0.65
   • Recommended for code snippets and documentation

Maintenance

1. Monthly vector store optimization:

   rm -rf ~/Documents/notes/.vector_store
   python main.py kb --reindex

2. Monitor similarity scores:

   • OpenAI: Use thresholds 0.60-0.85
   • Ollama: Use thresholds 0.50-0.70

Troubleshooting

1. Enable debug logging:

   # In config.yaml
   logging:
     level: "DEBUG"

2. Common fixes:

   # Clear vector store
   rm -rf ~/Documents/notes/.vector_store
   
   # Rebuild index
   python main.py kb --reindex
   
   # Check Ollama service
   curl http://localhost:11434/api/embeddings

Acknowledgments

This tool was created using:

• CodeLLama
• GPT-4o
• Claude 3.5 Sonnet
• Gemini 2 Flash
• Anthropic MCP
• Cursor
• aider.chat