Cellmage 0.1.3

No project description provided

🧙 CellMage ✨

Stop juggling browser tabs! Start casting LLM spells directly in your Jupyter Notebook cells!

Ever find yourself deep in a complex data analysis or coding session, only to hit a wall? You need a quick code snippet, an explanation, or just some creative brainstorming juice from your favorite Large Language Model. But ugh, that means breaking your flow, opening a new tab, copying, pasting, context-switching... the magic is gone.

Fear not, apprentice! CellMage is here to restore the enchantment! 🪄

CellMage is your personal LLM wizard, living right inside your Jupyter or IPython notebook. It provides powerful, intuitive magic commands to interact with LLMs without ever leaving your cell-based sanctuary.

What Sorcery is This? (Core Features)

CellMage isn't actual magic, but it feels pretty close:

• ✨ Intuitive Spellcasting: Use the %%llm cell magic. Write your prompt, run the cell, and poof – the LLM's response appears!
• 📜 Arcane History Scrolls: CellMage remembers your conversation. Even better, it detects when you re-run a cell and cleverly avoids duplicating history. Like turning back time, but less paradoxical!
• 🔮 Persona Grimoires: Define different LLM "personalities" (system prompts + configs) in simple Markdown files. Need a Python expert? A terse code generator? A rubber duck? Switch personas with a simple command!
• 🌍 Access Diverse Magical Fonts: Connect to OpenAI, Anthropic, Azure, and other OpenAI-compatible APIs.
• ⚡️ Live Conjuring: Watch the LLM weave its response character-by-character with built-in streaming support (the default!).
• 🪄 Ambient Enchantment (Optional): Feeling lazy? Use %llm_setup_forever to automatically treat any standard cell you run as a prompt! (Use %disable_llm_config_persistent to turn it off).
• 🧪 Precise Incantations: Override models, temperature, and other parameters on-the-fly for specific spells or configure instance-wide defaults. Use model aliases (like 'g4' for 'gpt-4') for quick access to your favorite models.
• 📝 Model Aliases: Define short aliases for your frequently used models in a cellmage_models.yml file or manage them through magic commands.
• GOTO Re-usable Spell Snippets: Inject content from local files (like code context or data samples) directly into the conversation history before casting your spell.
• 💰 Mana Tracking: Get a handy status bar after each call showing duration and estimated cost (because even magic has its price!).
• 📚 Detailed Spell Logs: Keep a record of your magical experiments with robust file logging.

The Installation Incantation 🪄

Ready to wield this power? Open your terminal and chant the sacred words:

# Basic installation
pip install cellmage

# For LangChain integration
pip install "cellmage[langchain]"

LLM Connection Options 🔌

CellMage provides a DirectLLMAdapter for connecting to LLM APIs:

DirectLLMAdapter (built-in with no additional dependencies):

• Connects directly to OpenAI-compatible APIs via standard HTTP requests
• Supports standard OpenAI API functionality including streaming
• Simple and reliable with minimal dependencies

Summoning Your First Spell (Quick Start) ⚡️

1. Load the Extension: In a notebook cell, run:

   %load_ext cellmage
   

2. Configure (Optional but Recommended): Tell CellMage where to find its power source. Set your API key and base URL (if needed) using environment variables (CELLMAGE_API_KEY, CELLMAGE_API_BASE) or the setup magic:

   # Example for a local OpenAI-compatible API server
   %llm_setup --api_base http://localhost:1234/v1 --default_model local-model
   

   Or for OpenAI:

   # Make sure CELLMAGE_API_KEY is set as an environment variable, or use:
   # %llm_setup --api_key "sk-..." --default_model gpt-4o
   

   (See "Connecting to Magical Fonts" below for more ways)

3. Cast Your Spell! In a new cell, use the %%llm magic:

   %%llm
   Explain the difference between a list and a tuple in Python like I'm five.
   

   Run the cell and behold the LLM's wisdom appearing below! ✨

Mastering the Arcane Arts (Core Concepts) 📖

Available Magic Commands

CellMage provides the following IPython magic commands:

• %%llm - Cell magic to send your prompt to the LLM
• %llm_config - Line magic to configure session state and manage resources
• %llm_config_persistent - Enable ambient mode (process regular cells as LLM prompts)
• %disable_llm_config_persistent - Disable ambient mode
• %%py - Execute a cell as normal Python code when ambient mode is active

For a comprehensive reference of all available magic commands and their arguments, see the IPython Magic Commands documentation.

The %%llm Cell Spell

This is your bread-and-butter. Anything in a cell marked with %%llm at the top gets sent as a prompt to the currently configured LLM.

%%llm --model gpt-4o --persona python_expert
Write a python function that takes a list of numbers and returns the sum,
but handle potential type errors gracefully.

Common arguments:

• -p, --persona: Temporarily use a different personality for this spell
• -m, --model: Temporarily use a different model for this spell
• -t, --temperature: Set temperature for this call
• --max-tokens: Set max_tokens for this call
• --no-stream: Disable streaming output just for this cell
• --no-history: Don't add this exchange to conversation history
• --param KEY VALUE: Set any other LLM param ad-hoc (e.g., --param top_p 0.9)

Configuring Your Session (%llm_config & %llm_config_persistent)

• %llm_config: Configure CellMage for the current session. Set defaults like your preferred model, API endpoints, persona folders, logging preferences, etc.

  %llm_config --model gpt-4o --persona coding_assistant --auto_save True --status
  

• %llm_config_persistent: Does the same as %llm_config, but also enables the "Ambient Enchantment" mode, treating subsequent non-magic cells as prompts. Great for pure chat sessions! Use %disable_llm_config_persistent to deactivate.

Model Aliases

CellMage supports defining short aliases for model names. You can:

1. Create a cellmage_models.yml file in your project directory:

# Model aliases
g4: gpt-4
g4t: gpt-4-turbo
c2: claude-2

2. Manage aliases through magic commands:

%llm_config --list-mappings  # Show current mappings
%llm_config --add-mapping g4 gpt-4  # Add new mapping
%llm_config --remove-mapping g4  # Remove mapping

Then use your aliases anywhere you'd use a model name:

%llm_config --model g4  # Use GPT-4
%%llm -m g4t  # Use GPT-4 Turbo for this cell

Ambient Enchantment (Auto-Processing)

Run %llm_config_persistent. Now, just type a prompt in a regular cell and run it! CellMage intercepts it and sends it to the LLM. Magic!

To run actual Python code while in ambient mode, use the %%py cell magic:

%%py
# This will run as normal Python code, not as an LLM prompt
x = 10
print(f"The value is {x}")

Remember to use %disable_llm_config_persistent when you want normal cell execution back.

Spell Snippets

Need to include the content of a file (like code context or data) in your prompt history before asking your question?

# Add content of my_code.py as a system message before the next %%llm call
%llm_setup --snippets my_code.py system

# Or add multiple files with a specific role
# %llm_setup --snippets file1.txt file2.json user

The snippet content will be added to the history for the next LLM call.

Multiple Persona and Snippet Folders

CellMage supports using personas and snippets from multiple directories, making it easier to organize resources by project or purpose:

1. Environment Variables: Set additional directories using comma-separated values:

   # Using environment variables
   export CELLMAGE_PERSONAS_DIRS=project_A/personas,project_B/personas
   export CELLMAGE_SNIPPETS_DIRS=project_A/snippets,project_B/snippets
   

2. Auto-discovery: CellMage automatically looks for personas and snippets in standard locations:

   • Root llm_personas and llm_snippets directories
   • notebooks/llm_personas, notebooks/llm_snippets
   • notebooks/examples and notebooks/tests subdirectories

3. Custom Loaders: For programmatic access to multiple directories:

   from cellmage.resources.file_loader import MultiFileLoader
   
   # Create a loader with multiple directories
   loader = MultiFileLoader(
       personas_dirs=["llm_personas", "project_A/personas"], 
       snippets_dirs=["llm_snippets", "project_A/snippets"]
   )
   
   # See available resources
   print(f"Available personas: {loader.list_personas()}")
   print(f"Available snippets: {loader.list_snippets()}")
   

Mana Tracking (Status Bar)

After each successful call, a small status bar appears showing: ✅ Success | ⏱️ Duration | 📥 Tokens In | 📤 Tokens Out | 🪙 Estimated Cost

(Cost estimation is basic and may vary wildly based on the model/provider)

Advanced Spellcraft

(Add examples here later for things like programmatic use, complex overrides, etc.)

# Example: Accessing the underlying object (if needed)
from cellmage import NotebookLLMMagics # Assuming this path
mage_instance = %magics_object NotebookLLMMagics
llm_helper = mage_instance.llm

# Now you can use methods directly
# llm_helper.set_override("temperature", 0.9)
# response = llm_helper.chat("This uses the override", personality_name="creative")
# llm_helper.remove_override("temperature")

Join the Coven? (Contributing)

Found a mischievous bug? Have an idea for a powerful new spell? Contributions are welcome! Please check the CONTRIBUTING.md file for guidelines.

The Fine Print (License)

CellMage is distributed under the MIT License. See LICENSE file for details. May cause spontaneous bursts of productivity and/or delight. Not responsible for accidentally summoning sentient code.

---

Happy Conjuring! ✨