Explore the Latest Open‑Source AI Projects: Llama 3, MaxKB, Phidata & RAGFlow

In this issue we recommend four open‑source AI projects:

Llama 3 large language model

MaxKB knowledge‑base Q&A system

Phidata framework for AI assistants

RAGFlow retrieval‑augmented generation engine

Llama 3 Large Language Model

Llama 3 is Meta’s latest open‑source LLM designed for individuals, creators, researchers and enterprises to responsibly experiment, innovate and scale ideas.

Key features compared with previous open‑source models:

Data: trained on more than seven times the dataset used for Llama 2.

Capability boost: improved reasoning and code generation.

Training efficiency: three times faster than Llama 2.

Model sizes: pretrained and instruction‑tuned variants ranging from 8 B to 70 B parameters.

Download & usage: provides model weights, tokenizer and quick‑start guide for local deployment.

Model parallelism: different parallelism settings required for each size.

License: weights and model are open to researchers and commercial entities to promote ethical AI progress.

GitHub: https://github.com/meta-llama/llama3

MaxKB Knowledge‑Base Q&A System

MaxKB is an LLM‑based knowledge‑base Q&A platform developed by 1Panel‑dev, aiming to become the “strongest brain” for enterprises. It has earned 2.9k GitHub stars.

Out‑of‑the‑box: supports document upload, web crawling, automatic text splitting and vectorisation, delivering a smooth interactive QA experience.

Seamless integration: can be embedded into third‑party systems without coding.

Multi‑model support: works with various mainstream LLMs, both private and cloud‑based.

GitHub: https://github.com/1Panel-dev/MaxKB

Phidata – Framework for Building AI Assistants

Phidata provides a framework to create AI assistants with memory, knowledge and tool integration, addressing LLM context limits and lack of action capability.

Memory: stores chat history in a database for long‑term conversations.

Knowledge: uses a vector database to supply contextual information to the LLM.

Tools: enables the LLM to perform actions such as API calls, sending emails, or querying databases.

GitHub: https://github.com/phidatahq/phidata

RAGFlow – Open‑Source Retrieval‑Augmented Generation Engine

RAGFlow, open‑sourced by infiniflow and starred by 5.2 K users, is a RAG engine that simplifies workflows for enterprises of any size through deep document understanding.

High‑quality I/O: extracts knowledge from complex, unstructured data.

Template‑based chunking: offers intelligent, explainable chunking options.

Reference‑based citations: reduces hallucinations with visual text chunks and manual intervention.

Heterogeneous data support: handles Word, PPT, Excel, TXT, images, scanned copies, structured data, web pages, etc.

Automated RAG workflow: provides configurable LLM and embedding models, multi‑stage retrieval and re‑ranking, and intuitive APIs for seamless business integration.

GitHub: https://github.com/infiniflow/ragflow