NetGent (Network Agent) is a powerful, functional API for designing multi-agent graph systems. Inspired by the functional APIs of LangGraph/LangChain and Keras/PyTorch, NetGent simplifies the creation of complex AI ecosystems.
The sequential processing workflow chains multiple agents for step-by-step processing, where each agent's output becomes the input for the next agent.
The parallel processing workflow executes multiple agents concurrently, where each agent processes the initial state independently.
The parallel processing workflow executes multiple agents concurrently, where each agent processes the initial state independently.
- Simplified Multi-Agent Systems: Create and manage multi-agent systems with ease.
- Seamless Integration: Combine generative AI (LLMs, VLMs, Audio models) with traditional neural networks.
- Flexible Architecture: Design your AI workflow using simple, intuitive building blocks.
- Unified Interface: Standardized
invoke(state: State) -> State
interface for all AI components. - Prompt Engineering Made Easy: Implement advanced prompt techniques without complexity.
from netgent import llm, response_average, sequential, parallel
# Simple invocation
result_state = llm(state)
# Average multiple responses
result_state = response_average(llm, evaluator=llm, state, k=5)
# Sequential processing
result_state = sequential(sequence=[llm_1, llm_2, llm_3], state)
# Parallel processing
result_state = parallel(agents=[llm_1, llm_2], state)
final_state = wrap_states(result_state)
pip install netgent
-
Develop NetGent Core functionality (v1.0.0)
- Implement various prompt types
- Enable different calling patterns (sequential, parallel, skip connections)
- Ensure robust support for complex multi-agent LLM architectures (text-based)
-
Expand multi-agent capabilities with multimodality (v2.0.0)
- Add vision agents
- Integrate speech-to-text agents
- Enhance multi-agent architectures with multimodal support
Note: All agents will adhere to the NetGent philosophy, maintaining a consistent invoke(state) -> state
interface.
NetGent's core philosophy is "Treat everything as an agent." By standardizing the interface for all AI components, NetGent enables developers to create truly AI-driven software with unprecedented ease and flexibility.
We welcome contributions! Please see our Contributing Guide for more details.
NetGent is released under the MIT License.
For detailed documentation, tutorials, and examples, visit our official documentation.
NetGent: Empowering developers to create the next generation of AI ecosystems.