BabyAGI
[!NOTE]
The original BabyAGI from March 2023 introduced task planning as a method for developing autonomous agents. This project has been archived and moved to the babyagi_archive repo (September 2024 snapshot).
[!CAUTION]
This is a framework built by Yohei who has never held a job as a developer. The purpose of this repo is to share ideas and spark discussion and for experienced devs to play with. Not meant for production use. Use with cautioun.
This newest BabyAGI is an experimental framework for a self-building autonomous agent. Earlier efforts to expand BabyAGI have made it clear that the optimal way to build a general autonomous agent is to build the simplest thing that can build itself.
Check out this introductory X/Twitter thread for a simple overview.
The core is a new function framework (functionz) for storing, managing, and executing functions from a database. It offers a graph-based structure for tracking imports, dependent functions, and authentication secrets, with automatic loading and comprehensive logging capabilities. Additionally, it comes with a dashboard for managing functions, running updates, and viewing logs.
Table of Contents
Quick Start
To quickly check out the dashboard and see how it works:
Install BabyAGI:
pip install babyagi
Import BabyAGI and load the dashboard:
import babyagi
if __name__ == "__main__":
app = babyagi.create_app('/dashboard')
app.run(host='0.0.0.0', port=8080)
Navigate to the dashboard:
Open your browser and go to http://localhost:8080/dashboard to access the BabyAGI dashboard.
Basic Usage
Start by importing babyagi and registering your functions. Here's how to register two functions, where one depends on the other:
import babyagi
# Register a simple function
@babyagi.register_function()
def world():
return "world"
# Register a function that depends on 'world'
@babyagi.register_function(dependencies=["world"])
def hello_world():
x = world()
return f"Hello {x}!"
# Execute the function
print(babyagi.hello_world()) # Output: Hello world!
if __name__ == "__main__":
app = babyagi.create_app('/dashboard')
app.run(host='0.0.0.0', port=8080)
Functions can be registered with metadata to enhance their capabilities and manage their relationships. Here's a more comprehensive example of function metadata, showing logical usage of all fields:
import babyagi
@babyagi.register_function(
imports=["math"],
dependencies=["circle_area"],
key_dependencies=["openai_api_key"],
metadata={
"description": "Calculates the volume of a cylinder using the circle_area function."
}
)
def cylinder_volume(radius, height):
import math
area = circle_area(radius)
return area * height
Available Metadata Fields:
imports: List of external libraries the function depends on.dependencies: List of other functions this function depends on.key_dependencies: List of secret keys required by the function.metadata["description"]: A description of what the function does.
Function Loading
In addition to using register_function, you can use load_function to load plugins or draft packs of functions. BabyAGI comes with built-in function packs, or you can load your own packs by pointing to the file path.
You can find available function packs in babyagi/functionz/packs.
Loading Custom Function Packs:
import babyagi
# Load your custom function pack
babyagi.load_functions("path/to/your/custom_functions.py")
This approach makes function building and management easier by organizing related functions into packs.
Key Dependencies
You can store key_dependencies directly from your code or manage them via the dashboard.
Storing Key Dependencies from Code:
import babyagi
# Add a secret key
babyagi.add_key_wrapper('openai_api_key', 'your_openai_api_key')
Adding Key Dependencies via Dashboard:
Navigate to the dashboard and use the add_key_wrapper feature to securely add your secret keys.
Execution Environment
BabyAGI automatically loads essential function packs and manages their dependencies, ensuring a seamless execution environment. Additionally, it logs all activities, including the relationships between functions, to provide comprehensive tracking of function executions and dependencies.
Log
BabyAGI implements a comprehensive logging system to track all function executions and their interactions. The logging mechanism ensures that every function call, including its inputs, outputs, execution time, and any errors, is recorded for monitoring and debugging purposes.
Key Logging Features:
Execution Tracking: Logs when a function starts and finishes execution, including the function name, arguments, keyword arguments, and execution time.
Error Logging: Captures and logs any errors that occur during function execution, providing detailed error messages for troubleshooting.
Dependency Management: Automatically resolves and logs dependencies between functions, ensuring that all required functions and libraries are loaded before execution.
Trigger Logging: Logs the execution of triggered functions, detailing which functions were triggered by others and their respective execution outcomes.
Comprehensive Records: Maintains a history of all function executions, enabling users to review past activities, understand function relationships, and analyze performance metrics.
How Triggers Work:
Triggers are mechanisms that allow certain functions to be automatically executed in response to specific events or actions within the system. For example, when a function is added or updated, a trigger can initiate the generation of a description for that function.
Triggers enhance the autonomy of BabyAGI by enabling automated workflows and reducing the need for manual intervention. However, it's essential to manage triggers carefully to avoid unintended recursive executions or conflicts between dependent functions.
Dashboard
The BabyAGI dashboard offers a user-friendly interface for managing functions, monitoring executions, and handling configurations. Key features include:
Function Management: Register, deregister, and update functions directly from the dashboard.
Dependency Visualization: View and manage dependencies between functions to understand their relationships.
Secret Key Management: Add and manage secret keys securely through the dashboard interface.
Logging and Monitoring: Access comprehensive logs of function executions, including inputs, outputs, and execution times.
Trigger Management: Set up triggers to automate function executions based on specific events or conditions.
Accessing the Dashboard:
After running your application, navigate to http://localhost:8080/dashboard to access the BabyAGI dashboard.
Pre-loaded Functions Summary
BabyAGI includes two pre-loaded function packs:
Default Functions (packs/default_functions.py):
- Function Execution: Run, add, update, or retrieve functions and versions.
- Key Management: Add and retrieve secret keys.
- Triggers: Add triggers to execute functions based on others.
- Logs: Retrieve logs with optional filters.
AI Functions (packs/ai_generator.py):
- AI Description & Embeddings: Auto-generate descriptions and embeddings for functions.
- Function Selection: Find or choose similar functions based on prompts.
Running a Self-Building Agent
BabyAGI includes two experimental self-building agents, showcasing how the framework can help a self-building coding agent leverage existing functions to write new ones.
This function first determines whether to use an existing function or generate new ones. If new functions are needed, it breaks them down into smaller reusable components and combines them into a final function.
Try this:
import babyagi
babyagi.add_key_wrapper('openai_api_key', os.environ['OPENAI_API_KEY'])
babyagi.load_functions("drafts/code_writing_functions")
babyagi.process_user_input("Grab today's score from ESPN and email it to [email protected]")
When you run this, you will see the functions being generated in the shell and new functions will be available in the dashboard once completed.
2. self_build in the self_build pack
This function takes a user description and generates X distinct tasks that a user might ask an AI assistant. Each task is processed by process_user_input, creating new functions if no existing ones suffice.
Try this:
import babyagi
babyagi.add_key_wrapper('openai_api_key', os.environ['OPENAI_API_KEY'])
babyagi.load_functions("drafts/code_writing_functions")
babyagi.load_functions("drafts/self_build")
babyagi.self_build("A sales person at an enterprise SaaS company.", 3)
This will generate 3 distinct tasks a salesperson might ask an AI assistant and create functions to handle those.
*The functions will be generated and stored in the dashboard, but note that the generated code is minimal and may need improvement.
Warning: These draft features are experimental concepts and may not function as intended. They require significant improvements and should be used with caution.
Contributing
Contributions are greatly appreciatedly, but candidly I have not been great at managing PRs. Please be patient as things will move slow while I am working on this alone (on nights and weekends). I may start by building a small core crew before collaborating with a larger group.
If you are a dev, investor, friend of open-source and interesting supporting AI work I do, please fill this form (I have a few fun initiatives coming up!)
License
BabyAGI is released under the MIT License. See the LICENSE file for more details.
BabyAGI 是一个实验性框架，用于构建可自我构建的自主智能体。它围绕函数框架 functionz 搭建，能够从数据库中存储、管理和执行函数。