This repository, AgentFlow, serves as a powerful Command Line Interface (CLI) tool for orchestrating multi-agent AI workflows. It integrates with large language models (LLMs) like Gemini and provides features for persistent memory, performance metrics, and extensibility.
- Multi-Agent Orchestration: Allows for the definition and execution of multiple AI agents that can interact with each other in a structured workflow.
- Persistent Memory System: Integrates a SQLite-based database to store conversational history and agent-specific states persistently.
- Performance Metrics (Token Usage): Tracks and reports token usage for interactions with LLMs.
- LLM Integration: Seamlessly integrates with LLMs, allowing their capabilities to be leveraged within agent workflows.
- Configurability: Allows users to configure agent behaviors, memory settings, and LLM interaction parameters.
- Extensibility: Agents are refactored into separate files to improve extensibility.
- Basic Unit Tests: Includes a basic unit test for the memory system.
- GitHub Actions: Basic CI workflow implemented for automated testing.
- Agent Tool Use (Placeholder): Includes a placeholder for agents to utilize external tools.
- CLI Commands for Memory Management: Provides command-line tools to manage conversation history.
-
Clone the repository:
git clone https://github.com/anchapin/AgentFlow.git cd AgentFlow -
Install dependencies:
npm install
To run the multi-agent conversation scenario:
node index.js-
View Conversation History:
node index.js --view-memory
-
Clear Conversation History:
node index.js --clear-memory
To run the basic unit tests:
npm test- User Interface: Develop a more interactive CLI or web-based interface.
- Advanced Metrics: Explore additional performance metrics beyond token usage, such as latency and cost estimation.
- Full Agent Tool Use: Fully implement agents' ability to use external tools.