MnemoAI/mnemo provides a framework for building AI agents, particularly those leveraging Retrieval-Augmented Generation (RAG). It simplifies the creation of custom AI assistants and tools, enabling developers to integrate various LLM providers and observability solutions. The primary benefit is the rapid development of intelligent agents capable of querying and synthesizing information from diverse data sources.

How It Works

Mnemo AI utilizes a tool-factory pattern to abstract the complexity of integrating with RAG pipelines and LLM services. Developers define agent tools, including RAG tools that query Mnemo corpora with optional metadata filtering, and other specialized tools (e.g., for finance, legal, or database interaction). These tools are then composed into an Agent object, which can be configured with different agent types (ReAct, LLMCompiler, etc.) and LLM providers. The framework supports asynchronous operations and streaming responses.

Quick Start & Requirements

• Install: pip install mnemo-agentic
• Prerequisites: Requires API keys for Mnemo services (e.g., MNEMO_API_KEY, MNEMO_CUSTOMER_ID, MNEMO_CORPUS_ID) and potentially other LLM providers, set as environment variables.
• Quick Start: Initialize MnemoToolFactory, create RAG or search tools using create_rag_tool() or create_search_tool(), define an Agent with tools and custom instructions, and run using agent.chat().
• Docs: https://github.com/mnemo-agentic/mnemo

Highlighted Details

• Supports multiple agent types: ReAct, OpenAIAgent, LATS, LLMCompiler.
• Integrates with numerous LLM providers: OpenAI, DeepSeek, Anthropic, Gemini, GROQ, Together.AI, Cohere, Bedrock, Fireworks.
• Built-in observability with Arize Phoenix.
• Extensive pre-built tools, including RAG, search, database interaction, and integrations with services like Tavily, EXA.AI, and Google tools.
• Supports metadata filtering for RAG queries, including conditional filtering and fixed filters.

Maintenance & Community

• Community: Twitter: @mnemo_ai, GitHub: MnemoAI
• Contributing: CONTRIBUTING.md

Licensing & Compatibility

• License: Apache 2.0 License.
• Compatibility: Permissive license suitable for commercial use and integration with closed-source applications.

Limitations & Caveats

The framework requires specific Mnemo API keys and potentially other third-party API keys for full functionality, which may incur costs. While it supports various LLM providers, optimal performance and compatibility may depend on the chosen provider and model.