grami-ai 0.3.129

A dynamic and flexible AI agent framework for building intelligent, multi-modal AI agents

GRAMI-AI: Dynamic AI Agent Framework

Overview

GRAMI-AI is a cutting-edge, async-first AI agent framework designed to solve complex computational challenges through intelligent, collaborative agent interactions. Built with unprecedented flexibility, this library empowers developers to create sophisticated, context-aware AI systems that can adapt, learn, and collaborate across diverse domains.

Key Features

• Async AI Agent Creation
• Multi-LLM Support (Gemini, OpenAI, Anthropic, Ollama)
• Extensible Tool Ecosystem
• Multiple Communication Interfaces
• Flexible Memory Management
• Secure and Scalable Architecture

Installation

Using pip

pip install grami-ai

From Source

git clone https://github.com/YAFATEK/grami-ai.git
cd grami-ai
pip install -e .

Quick Start

Basic Async Agent Creation

import asyncio
from grami.agent import AsyncAgent
from grami.providers.gemini_provider import GeminiProvider

async def main():
    # Initialize a Gemini-powered Async Agent
    agent = AsyncAgent(
        name="AssistantAI",
        llm=GeminiProvider(api_key="YOUR_API_KEY"),
        system_instructions="You are a helpful digital assistant."
    )

    # Send an async message
    response = await agent.send_message("Hello, how can you help me today?")
    print(response)

    # Stream a response
    async for token in agent.stream_message("Tell me a story"):
        print(token, end='', flush=True)

asyncio.run(main())

Example Configurations

1. Async Agent with Memory and Streaming

from grami.agent import AsyncAgent
from grami.providers.gemini_provider import GeminiProvider
from grami.memory.lru import LRUMemory

agent = AsyncAgent(
    name="MemoryStreamingAgent",
    llm=provider,
    memory=LRUMemory(capacity=100),
    system_instructions="You are a storyteller."
)

2. Async Agent without Memory

agent = AsyncAgent(
    name="NoMemoryAgent",
    llm=provider,
    memory=None,
    system_instructions="You are a concise assistant."
)

3. Async Agent with Streaming Disabled

response = await agent.send_message("Tell me about AI")

4. Async Agent with Streaming Enabled

async for token in agent.stream_message("Explain quantum computing"):
    print(token, end='', flush=True)

Memory Providers

Grami AI supports multiple memory providers to suit different use cases:

1. LRU Memory: A local, in-memory cache with a configurable capacity.

   from grami.memory import LRUMemory
   
   # Initialize with default 100-item capacity
   memory = LRUMemory(capacity=50)
   

2. Redis Memory: A distributed memory provider using Redis for scalable, shared memory storage.

   from grami.memory import RedisMemory
   
   # Initialize with custom Redis configuration
   memory = RedisMemory(
       host='localhost',  # Redis server host
       port=6379,         # Redis server port
       capacity=100,      # Maximum number of items
       provider_id='my_agent'  # Optional provider identifier
   )
   
   # Store memory items
   await memory.store('user_query', 'What is AI?')
   await memory.store('agent_response', 'AI is...')
   
   # Retrieve memory items
   query = await memory.retrieve('user_query')
   
   # List memory contents
   contents = await memory.list_contents()
   
   # Get recent items
   recent_items = await memory.get_recent_items(limit=5)
   
   # Clear memory
   await memory.clear()
   

   Redis Memory Prerequisites

   • Install Redis server locally or use a cloud-hosted Redis instance
   • Ensure network accessibility to Redis server
   • Install additional dependencies:

     pip install grami-ai[redis]
     

   Redis Memory Configuration Options

   • host: Redis server hostname (default: 'localhost')
   • port: Redis server port (default: 6379)
   • db: Redis database number (default: 0)
   • capacity: Maximum memory items (default: 100)
   • provider_id: Unique memory namespace identifier

   Best Practices

   • Use unique provider_id for different conversations
   • Set appropriate capacity based on memory requirements
   • Handle potential network or connection errors
   • Consider Redis persistence settings for data durability

Memory Usage with LLM Providers

Memory providers can be seamlessly integrated with LLM providers:

# Example with Gemini Provider
gemini_provider = GeminiProvider(
    model_name='gemini-pro',
    memory=memory  # Use either LRUMemory or RedisMemory
)

Development Checklist

Core Framework Design

• Implement AsyncAgent base class with dynamic configuration
• Create flexible system instruction definition mechanism
• Design abstract LLM provider interface
• Develop dynamic role and persona assignment system
• Comprehensive async example configurations
  • Memory with streaming
  • Memory without streaming
  • No memory with streaming
  • No memory without streaming
• Implement multi-modal agent capabilities (text, image, video)

LLM Provider Abstraction

• Unified interface for diverse LLM providers
  • Google Gemini integration (start_chat(), send_message())
    • Basic message sending
    • Streaming support
    • Memory integration
  • OpenAI ChatGPT integration
    • Basic message sending
    • Streaming implementation
    • Memory support
  • Anthropic Claude integration
  • Ollama local LLM support
• Standardize function/tool calling across providers
• Dynamic prompt engineering support
• Provider-specific configuration handling

Communication Interfaces

• WebSocket real-time communication
• REST API endpoint design
• Kafka inter-agent communication
• gRPC support
• Event-driven agent notification system
• Secure communication protocols

Memory and State Management

• Pluggable memory providers
  • In-memory state storage
  • Redis distributed memory
  • DynamoDB scalable storage
  • S3 content storage
• Conversation and task history tracking
• Global state management for agent crews
• Persistent task and interaction logs
• Advanced memory indexing
• Memory compression techniques

Tool and Function Ecosystem

• Extensible tool integration framework
• Default utility tools
  • Kafka message publisher
  • Web search utility
  • Content analysis tool
• Provider-specific function calling support
• Community tool marketplace
• Easy custom tool development

Agent Crew Collaboration

• Inter-agent communication protocol
• Workflow and task delegation mechanisms
• Approval and review workflows
• Notification and escalation systems
• Dynamic team composition
• Shared context and memory management

Use Case Implementations

• Digital Agency workflow template
  • Growth Manager agent
  • Content Creator agent
  • Trend Researcher agent
  • Media Creation agent
• Customer interaction management
• Approval and revision cycles

Security and Compliance

• Secure credential management
• Role-based access control
• Audit logging
• Compliance with data protection regulations

Performance and Scalability

• Async-first design
• Horizontal scaling support
• Performance benchmarking
• Resource optimization

Testing and Quality

• Comprehensive unit testing
• Integration testing for agent interactions
• Mocking frameworks for LLM providers
• Continuous integration setup

Documentation and Community

• Detailed API documentation
• Comprehensive developer guides
• Example use case implementations
• Contribution guidelines
• Community tool submission process
• Regular maintenance and updates

Future Roadmap

• Payment integration solutions
• Advanced agent collaboration patterns
• Specialized industry-specific agents
• Enhanced security features
• Extended provider support

Contributing

Contributions are welcome! Please check our GitHub repository for guidelines.

Support

• Email: [e-mail]
• GitHub: GRAMI-AI Issues

---

2024 YAFATEK. All Rights Reserved.