https://github.com/msoedov/meta_loop
An agent-optimization framework ✨
https://github.com/msoedov/meta_loop
agent-framework agent-frameworks agentic-workflow crewai pydantic-ai
Last synced: about 1 month ago
JSON representation
An agent-optimization framework ✨
- Host: GitHub
- URL: https://github.com/msoedov/meta_loop
- Owner: msoedov
- License: mit
- Created: 2025-02-23T22:03:33.000Z (about 2 months ago)
- Default Branch: main
- Last Pushed: 2025-03-13T11:42:15.000Z (about 1 month ago)
- Last Synced: 2025-03-13T12:23:13.139Z (about 1 month ago)
- Topics: agent-framework, agent-frameworks, agentic-workflow, crewai, pydantic-ai
- Language: Python
- Homepage:
- Size: 1.76 MB
- Stars: 1
- Watchers: 1
- Forks: 0
- Open Issues: 1
-
Metadata Files:
- Readme: Readme.md
- License: LICENSE
Awesome Lists containing this project
README
Meta Loop
An automatic agent optimization framework for generating, evaluating, and selecting top-performing agents.
Define your task, and let Meta Loop discover the perfect solution.
Explore the Docs » · Report a Bug »
Meta Agent is a powerful, imperative, define-by-run framework designed to simplify agent development. It generates 16 candidate "probes" with varied configurations, evaluates them against customizable metrics, and selects the best-performing agent for your task. Whether you're analyzing data, predicting outcomes, or automating workflows, Meta Loop optimizes the process for you.
## ✨ Features
- Automated Agent Generation
- Creates 16 unique agent "probes" with diverse configurations tailored to your task.
- Customizable Evaluation
- Scores probes using built-in metrics (e.g., accuracy, speed) or your own custom evaluation function.
- Framework Flexibility
- Built with Pydantic-AI by default, with seamless support for frameworks like AutoGen or CrewAI.
- Extensible Design
- Easily adapt agent roles, tools, and evaluation criteria to suit your specific needs.
## 📦 Installation
Install Meta Agent with a single command:
```shell
pip install meta_loop
```
## 🚀 Quick Start
Get started in just a few lines of code. Below are examples to showcase Meta Agent’s capabilities.
Basic Usage
Generate an agent to analyze customer reviews and predict sentiment:
```python
import meta_loop
# Build and optimize an agent
best_agent = meta_loop.build_agent(
instruction="Create an agent to analyze customer reviews and predict sentiment.",
probe_count=16,
framework="crewai"
)
# View the selected agent's details
print(best_agent.details)
```
### Custom Evaluation Function
Define your own scoring logic to evaluate probes:
```python
import meta_loop
import random
# Custom evaluation function
def custom_eval(trial: meta_loop.Trial) -> float:
return random.random() # Replace with your own metric
# Build an agent with custom evaluation
best_agent = meta_loop.build_agent(
instruction="Create an agent to analyze customer reviews and predict sentiment.",
probe_count=16,
framework="crewai",
eval_fn=custom_eval
)
print(best_agent.details)
```
### Using a Test Dataset
Provide a dataset to evaluate agents against specific inputs and expected outputs:
```python
import meta_loop
# Build an agent with a test dataset
best_agent = meta_loop.build_agent(
instruction="Create an agent to analyze customer reviews and predict sentiment.",
probe_count=16,
framework="crewai",
test_dataset=meta_loop.dataset(
("Great product, love it!", 0.9), # (input, expected_score)
("Terrible service, very disappointed.", 0.2),
("It's okay, nothing special.", 0.5)
)
)
print(best_agent.details)
```
## 🛠️ How It Works
- Define Your Task: Provide a task description (e.g., "analyze customer reviews").
- Generate Probes: Meta Agent creates 16 agent configurations with varying parameters.
- Evaluate Performance: Each probe is scored based on your chosen metrics or dataset.
- Select the Best: The top-performing agent is returned, ready for use.
## 📚 Documentation
For more details, check out the official documentation (coming soon!).
## 🤝 Contributing
We welcome contributions! Please see our contribution guidelines and feel free to submit issues or pull requests.
## 📬 Get in Touch
Have questions? Join our community or reach out:
- Discord (coming soon!)
- GitHub Issues