pathwaycom/llm-app
Ready-to-run cloud templates for building real-time RAG, AI pipelines, and enterprise search applications that synchronize with various live data sources.
Awesome Infra for AI › Vector Databases & Retrieval Infrastructure
FlowElement-xinliuyuansu/m_flow
M-flow is a Python retrieval engine that treats a knowledge graph as the scoring mechanism rather than as a way to organize context. Knowledge is stored in a four-level cone graph: an Episode is a bounded semantic focus such as an incident or decision process, a Facet is one topical cross-section of that Episode, a FacetPoint is an atomic assertion derived from a Facet, and an Entity is a named person, tool or metric linked across all Episodes. When a query arrives, vector search first casts a wide net across all granularities to find entry points, so a precise cue anchors on a fine-grained FacetPoint while a broader theme anchors on a Facet or an Episode summary. The graph then takes over: evidence propagates along typed, semantically weighted edges, each hop adding cost, and every knowledge unit is scored by the strongest chain of reasoning connecting it to the query. Results are returned as Episode bundles, each containing the Episode together with its Facets and FacetPoints, which a downstream LLM uses to compose the final answer. The design argument is that similarity — proximity in representation space — and relevance — the existence of a coherent evidence structure connecting query to answer — are different properties, and that path-cost propagation captures the second where pure vector ranking captures only the first. Association is modeled as controlled propagation rather than a random graph walk, since only low-cost coherent paths stay competitive as the semantic field expands. The project targets Python 3.10 through 3.13, ships under Apache 2.0 with a large test suite, and publishes retrieval-architecture documentation covering the path-cost mechanism. It is aimed at teams building RAG systems and long-term agent memory where multi-hop reasoning over structured context matters more than nearest-neighbour chunk matching.
https://github.com/FlowElement-xinliuyuansu/m_flow
Also in Vector Databases & Retrieval Infrastructure
run-llama/llama_index
LlamaIndex is an open-source data framework for building LLM applications by connecting custom data sources to large language models, focusing on data ingestion, indexing, and retrieval augmented g...
milvus-io/milvus
Milvus is a high-performance, cloud-native vector database designed for scalable vector Approximate Nearest Neighbor (ANN) search, efficiently organizing and searching vast amounts of unstructured ...
VectifyAI/PageIndex
PageIndex is a vectorless, reasoning-based RAG system that builds hierarchical tree indexes from documents and uses LLMs to reason over them for context-aware retrieval.
qdrant/qdrant
Qdrant is an open-source, high-performance vector similarity search engine and vector database designed specifically for AI applications, enabling fast storage, search, and management of vectors wi...
Tencent/WeKnora
WeKnora is an open-source, LLM-powered knowledge framework for enterprise document understanding, semantic retrieval, and autonomous reasoning, featuring RAG, ReAct agents, and an auto-maintaining ...
topoteretes/cognee
Cognee is an open-source AI memory platform that provides AI agents with persistent long-term memory through a self-hosted knowledge graph, combining vector embeddings and graph reasoning.
RyanCodrai/turbovec
TurboVec is a Rust-based approximate nearest neighbor (ANN) vector index with Python bindings, built on Google Research's TurboQuant algorithm for efficient, memory-optimized vector similarity search.