Check out the announcement video:\nhttps://youtu.be/dsesHoTXyk0
GraphRAG is a research project from Microsoft exploring the use of knowledge graphs and large language models for enhanced retrieval augmented generation. It is an end-to-end system for richly understanding text-heavy datasets by combining text extraction, network analysis, LLM prompting, and summarization.
For more details on GraphRAG check out aka.ms/graphrag
Try out the Python code on your own machine: https://github.com/microsoft/graphrag
Easily deploy GraphRAG in Azure: https://github.com/Azure-Samples/graphrag-accelerator
Leave a comment below for what you want to build with GraphRAG!"},"title":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["graphrag"],"value":"GraphRAG is now on GitHub"},"url":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["graphrag"],"value":"https://www.microsoft.com/en-us/research/blog/graphrag-new-tool-for-complex-data-discovery-now-on-github/"}},"_tags":["story","author_alexchaomander","story_40857174"],"author":"alexchaomander","children":[40857175,40858786,40858791,40858860,40859158,40859175,40859280,40859347,40859451,40859530,40859901,40860034,40860124,40860252,40860346,40860604,40861155,40862374,40862460,40877141,40880529],"created_at":"2024-07-02T14:41:19Z","created_at_i":1719931279,"num_comments":49,"objectID":"40857174","points":282,"story_id":40857174,"story_text":"The team at Microsoft is pleased to announce that GraphRAG is now available in open-source!
Check out the announcement video:\nhttps://youtu.be/dsesHoTXyk0
GraphRAG is a research project from Microsoft exploring the use of knowledge graphs and large language models for enhanced retrieval augmented generation. It is an end-to-end system for richly understanding text-heavy datasets by combining text extraction, network analysis, LLM prompting, and summarization.
For more details on GraphRAG check out aka.ms/graphrag
Try out the Python code on your own machine: https://github.com/microsoft/graphrag
Easily deploy GraphRAG in Azure: https://github.com/Azure-Samples/graphrag-accelerator
Leave a comment below for what you want to build with GraphRAG!","title":"GraphRAG is now on GitHub","updated_at":"2025-10-15T15:03:21Z","url":"https://www.microsoft.com/en-us/research/blog/graphrag-new-tool-for-complex-data-discovery-now-on-github/"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"jinqueeny"},"title":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["graphrag"],"value":"Autoflow, a Graph RAG based and conversational knowledge base tool"},"url":{"matchLevel":"none","matchedWords":[],"value":"https://github.com/pingcap/autoflow"}},"_tags":["story","author_jinqueeny","story_42210689"],"author":"jinqueeny","children":[42211461,42211677,42211956,42211969,42212052,42212270,42213116,42214398,42216805],"created_at":"2024-11-22T02:42:14Z","created_at_i":1732243334,"num_comments":36,"objectID":"42210689","points":280,"story_id":42210689,"title":"Autoflow, a Graph RAG based and conversational knowledge base tool","updated_at":"2026-02-14T12:49:52Z","url":"https://github.com/pingcap/autoflow"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"taikon"},"title":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["graphrag"],"value":"KAG \u2013 Knowledge Graph RAG Framework"},"url":{"matchLevel":"none","matchedWords":[],"value":"https://github.com/OpenSPG/KAG"}},"_tags":["story","author_taikon","story_42545986"],"author":"taikon","children":[42546617,42546890,42546980,42547133,42547305,42547671,42547684,42547795,42548043,42548531,42548901,42551748,42559317,42572747],"created_at":"2024-12-30T02:55:58Z","created_at_i":1735527358,"num_comments":80,"objectID":"42545986","points":230,"story_id":42545986,"title":"KAG \u2013 Knowledge Graph RAG Framework","updated_at":"2026-04-28T00:48:57Z","url":"https://github.com/OpenSPG/KAG"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"sdht0"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["graphrag"],"value":"We show the potential of modern, embedded graph databases in the browser by demonstrating a fully in-browser chatbot that can perform Graph RAG using Kuzu (the graph database we're building) and WebLLM, a popular in-browser inference engine for LLMs. The post retrieves from the graph via a Text-to-Cypher pipeline that translates a user question into a Cypher query, and the LLM uses the retrieved results to synthesize a response. As LLMs get better, and WebGPU and Wasm64 become more widely adopted, we expect to be able to do more and more in the browser in combination with LLMs, so a lot of the performance limitations we see currently may not be as much of a problem in the future.
We will soon also be releasing a vector index as part of Kuzu that you can also use in the browser to build traditional RAG or Graph RAG that retrieves from both vectors and graphs. The system has come a long way since we open sourced it about 2 years ago, so please give us feedback about how it can be more useful!"},"title":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["graphrag"],"value":"Show HN: In-Browser Graph RAG with Kuzu-WASM and WebLLM"},"url":{"matchLevel":"none","matchedWords":[],"value":"https://blog.kuzudb.com/post/kuzu-wasm-rag/"}},"_tags":["story","author_sdht0","story_43321523","show_hn"],"author":"sdht0","children":[43321898,43322008,43322111,43322298,43324025,43325015,43325942,43326077,43327239,43328647],"created_at":"2025-03-10T15:12:57Z","created_at_i":1741619577,"num_comments":29,"objectID":"43321523","points":158,"story_id":43321523,"story_text":"We show the potential of modern, embedded graph databases in the browser by demonstrating a fully in-browser chatbot that can perform Graph RAG using Kuzu (the graph database we're building) and WebLLM, a popular in-browser inference engine for LLMs. The post retrieves from the graph via a Text-to-Cypher pipeline that translates a user question into a Cypher query, and the LLM uses the retrieved results to synthesize a response. As LLMs get better, and WebGPU and Wasm64 become more widely adopted, we expect to be able to do more and more in the browser in combination with LLMs, so a lot of the performance limitations we see currently may not be as much of a problem in the future.
We will soon also be releasing a vector index as part of Kuzu that you can also use in the browser to build traditional RAG or Graph RAG that retrieves from both vectors and graphs. The system has come a long way since we open sourced it about 2 years ago, so please give us feedback about how it can be more useful!","title":"Show HN: In-Browser Graph RAG with Kuzu-WASM and WebLLM","updated_at":"2026-07-15T04:15:49Z","url":"https://blog.kuzudb.com/post/kuzu-wasm-rag/"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"ktyptorio"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["graphrag"],"value":"Hi HN,
I have been working with regulation-heavy documents lately, and one thing kept bothering me. Flat RAG pipelines often fail to retrieve related articles together, even when they are clearly connected through references, definitions, or clauses.
After trying several RAG setups, I subjectively felt that GraphRAG was a better mental model for this kind of data. The Microsoft GraphRAG paper and reference implementation were helpful starting points. However, in practice, I found one recurring friction point: graph storage and vector indexing are usually handled by separate systems, which felt unnecessarily heavy for short-lived analysis tasks.
To explore this tradeoff, I built GibRAM (Graph in-buffer Retrieval and Associative Memory). It is an experimental, in-memory GraphRAG runtime where entities, relationships, text units, and embeddings live side by side in a single process.
GibRAM is intentionally ephemeral. It is designed for exploratory tasks like summarization or conversational querying over a bounded document set. Data lives in memory, scoped by session, and is automatically cleaned up via TTL. There are no durability guarantees, and recomputation is considered cheaper than persistence for the intended use cases.
This is not a database and not a production-ready system. It is a casual project, largely vibe-coded, meant to explore what GraphRAG looks like when memory is the primary constraint instead of storage. Technical debt exists, and many tradeoffs are explicit.
The project is open source, and I would really appreciate feedback, especially from people working on RAG, search infrastructure, or graph-based retrieval.
GitHub: https://github.com/gibram-io/gibram
Happy to answer questions or hear why this approach might be flawed."},"title":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["graphrag"],"value":"Show HN: GibRAM an in-memory ephemeral GraphRAG runtime for retrieval"},"url":{"matchLevel":"none","matchedWords":[],"value":"https://github.com/gibram-io/gibram"}},"_tags":["story","author_ktyptorio","story_46665393","show_hn"],"author":"ktyptorio","children":[46665823,46666886,46667099,46670077,46670115,46670156,46688495],"created_at":"2026-01-18T06:47:17Z","created_at_i":1768718837,"num_comments":9,"objectID":"46665393","points":60,"story_id":46665393,"story_text":"Hi HN,
I have been working with regulation-heavy documents lately, and one thing kept bothering me. Flat RAG pipelines often fail to retrieve related articles together, even when they are clearly connected through references, definitions, or clauses.
After trying several RAG setups, I subjectively felt that GraphRAG was a better mental model for this kind of data. The Microsoft GraphRAG paper and reference implementation were helpful starting points. However, in practice, I found one recurring friction point: graph storage and vector indexing are usually handled by separate systems, which felt unnecessarily heavy for short-lived analysis tasks.
To explore this tradeoff, I built GibRAM (Graph in-buffer Retrieval and Associative Memory). It is an experimental, in-memory GraphRAG runtime where entities, relationships, text units, and embeddings live side by side in a single process.
GibRAM is intentionally ephemeral. It is designed for exploratory tasks like summarization or conversational querying over a bounded document set. Data lives in memory, scoped by session, and is automatically cleaned up via TTL. There are no durability guarantees, and recomputation is considered cheaper than persistence for the intended use cases.
This is not a database and not a production-ready system. It is a casual project, largely vibe-coded, meant to explore what GraphRAG looks like when memory is the primary constraint instead of storage. Technical debt exists, and many tradeoffs are explicit.
The project is open source, and I would really appreciate feedback, especially from people working on RAG, search infrastructure, or graph-based retrieval.
GitHub: https://github.com/gibram-io/gibram
Happy to answer questions or hear why this approach might be flawed.","title":"Show HN: GibRAM an in-memory ephemeral GraphRAG runtime for retrieval","updated_at":"2026-03-05T23:27:11Z","url":"https://github.com/gibram-io/gibram"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"earayu"},"title":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["graphrag"],"value":"ApeRAG: Production-ready GraphRAG with multi-modal indexing and K8s deployment"},"url":{"matchLevel":"none","matchedWords":[],"value":"https://github.com/apecloud/ApeRAG"}},"_tags":["story","author_earayu","story_45165751"],"author":"earayu","children":[45216204,45216244,45216725,45217287],"created_at":"2025-09-08T08:06:43Z","created_at_i":1757318803,"num_comments":14,"objectID":"45165751","points":32,"story_id":45165751,"title":"ApeRAG: Production-ready GraphRAG with multi-modal indexing and K8s deployment","updated_at":"2026-03-20T04:32:51Z","url":"https://github.com/apecloud/ApeRAG"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"gusye"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["graphrag"],"value":"I made a simple GraphRAG call nano-graphrag.\nThe reason is I try to hack the official implementation released by Microsoft but that version is very hard to read/hack. \nThis algorithm should not be implemented that annoying, I think. So I made a simpler one.
It's about 800-900 lines of Python, and it's portable."},"title":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["graphrag"],"value":"Show HN: A Simple GraphRAG Implementation"},"url":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["graphrag"],"value":"https://github.com/gusye1234/nano-graphrag"}},"_tags":["story","author_gusye","story_41262991","show_hn"],"author":"gusye","children":[41265502,41270474,41281119],"created_at":"2024-08-16T03:48:02Z","created_at_i":1723780082,"num_comments":4,"objectID":"41262991","points":30,"story_id":41262991,"story_text":"I made a simple GraphRAG call nano-graphrag.\nThe reason is I try to hack the official implementation released by Microsoft but that version is very hard to read/hack. \nThis algorithm should not be implemented that annoying, I think. So I made a simpler one.
It's about 800-900 lines of Python, and it's portable.","title":"Show HN: A Simple GraphRAG Implementation","updated_at":"2024-10-23T18:24:05Z","url":"https://github.com/gusye1234/nano-graphrag"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"dmezzetti"},"title":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["graphrag"],"value":"Graph RAG for Wikipedia and ArXiv"},"url":{"matchLevel":"none","matchedWords":[],"value":"https://neuml.hashnode.dev/generate-knowledge-with-semantic-graphs-and-rag"}},"_tags":["story","author_dmezzetti","story_39357917"],"author":"dmezzetti","created_at":"2024-02-13T14:29:59Z","created_at_i":1707834599,"num_comments":0,"objectID":"39357917","points":21,"story_id":39357917,"title":"Graph RAG for Wikipedia and ArXiv","updated_at":"2024-09-20T16:26:48Z","url":"https://neuml.hashnode.dev/generate-knowledge-with-semantic-graphs-and-rag"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"Anon84"},"title":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["graphrag"],"value":"GraphRAG with LangChain and Neo4j"},"url":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["graphrag"],"value":"https://valentinaalto.medium.com/introducing-graphrag-with-langchain-and-neo4j-90446df17c1e"}},"_tags":["story","author_Anon84","story_40190013"],"author":"Anon84","children":[40193493,40194317,40197572],"created_at":"2024-04-28T17:05:52Z","created_at_i":1714323952,"num_comments":3,"objectID":"40190013","points":14,"story_id":40190013,"title":"GraphRAG with LangChain and Neo4j","updated_at":"2024-09-20T16:57:12Z","url":"https://valentinaalto.medium.com/introducing-graphrag-with-langchain-and-neo4j-90446df17c1e"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"acossta"},"title":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["graphrag"],"value":"GraphRAG: Unlocking LLM discovery on narrative private data"},"url":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["graphrag"],"value":"https://www.microsoft.com/en-us/research/blog/graphrag-unlocking-llm-discovery-on-narrative-private-data/"}},"_tags":["story","author_acossta","story_39591774"],"author":"acossta","created_at":"2024-03-04T15:46:10Z","created_at_i":1709567170,"num_comments":0,"objectID":"39591774","points":14,"story_id":39591774,"title":"GraphRAG: Unlocking LLM discovery on narrative private data","updated_at":"2024-09-20T16:34:36Z","url":"https://www.microsoft.com/en-us/research/blog/graphrag-unlocking-llm-discovery-on-narrative-private-data/"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"cbrizz00"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["graphrag"],"value":"I've been seeing a lot of buzz around GraphRAG and its potential, but many implementations seem to encounter issues, often with Neo4j. Has anyone managed to set up a reliable and cost-effective GraphRAG system in production? I'm curious about real-world experiences and practical solutions."},"title":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["graphrag"],"value":"Is anyone deploying GraphRAG in prod?"}},"_tags":["story","author_cbrizz00","story_41597587","ask_hn"],"author":"cbrizz00","children":[41598438,41599279],"created_at":"2024-09-20T00:09:11Z","created_at_i":1726790951,"num_comments":5,"objectID":"41597587","points":13,"story_id":41597587,"story_text":"I've been seeing a lot of buzz around GraphRAG and its potential, but many implementations seem to encounter issues, often with Neo4j. Has anyone managed to set up a reliable and cost-effective GraphRAG system in production? I'm curious about real-world experiences and practical solutions.","title":"Is anyone deploying GraphRAG in prod?","updated_at":"2026-02-18T12:05:06Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"dmezzetti"},"story_text":{"matchLevel":"none","matchedWords":[],"value":"Hello. This is an easy-to-use application for exploring your own data with retrieval augmented generation (RAG) backed by txtai.
txtai is an all-in-one embeddings database for semantic search, LLM orchestration and language model workflows. txtai has a feature to automatically create knowledge graphs using semantic similarity. This enables running Graph RAG queries with path traversals. This RAG application generates a visual network to illustrate the path traversals and help understand the context from which answers are generated from.
Embeddings databases are used as the knowledge store. The application can start with a blank database or an existing one such as Wikipedia. In both cases, new data can be added. This enables augmenting a large data source with new/custom information.
Adding new data is done with the textractor pipeline. This pipeline can extract content from documents (PDF, Word, etc) along with websites. The website extraction logic detects the likely sections with main content removing noisy sections such as headers and sidebars. This helps improve the overall RAG accuracy.
This RAG application is open source with the code available here: https://github.com/neuml/rag"},"title":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["graphrag"],"value":"Show HN: Open-source GraphRAG Application with your own data"},"url":{"matchLevel":"none","matchedWords":[],"value":"https://hub.docker.com/r/neuml/rag"}},"_tags":["story","author_dmezzetti","story_41194825","show_hn"],"author":"dmezzetti","created_at":"2024-08-08T18:41:24Z","created_at_i":1723142484,"num_comments":0,"objectID":"41194825","points":12,"story_id":41194825,"story_text":"Hello. This is an easy-to-use application for exploring your own data with retrieval augmented generation (RAG) backed by txtai.
txtai is an all-in-one embeddings database for semantic search, LLM orchestration and language model workflows. txtai has a feature to automatically create knowledge graphs using semantic similarity. This enables running Graph RAG queries with path traversals. This RAG application generates a visual network to illustrate the path traversals and help understand the context from which answers are generated from.
Embeddings databases are used as the knowledge store. The application can start with a blank database or an existing one such as Wikipedia. In both cases, new data can be added. This enables augmenting a large data source with new/custom information.
Adding new data is done with the textractor pipeline. This pipeline can extract content from documents (PDF, Word, etc) along with websites. The website extraction logic detects the likely sections with main content removing noisy sections such as headers and sidebars. This helps improve the overall RAG accuracy.
This RAG application is open source with the code available here: https://github.com/neuml/rag","title":"Show HN: Open-source GraphRAG Application with your own data","updated_at":"2025-04-07T12:56:50Z","url":"https://hub.docker.com/r/neuml/rag"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"kaifahmad1"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["graphrag"],"value":"Hi HN,
I\u2019m sharing Semantica, an MIT-licensed open-source framework for building semantic layers and knowledge engineering systems for AI.
Many RAG and agent systems fail not due to model quality, but due to the semantic gap \u2014 unstructured, inconsistent data without explicit entities, rules, or relationships. Vector-only approaches often hallucinate or fail silently under real-world data.
Semantica focuses on transforming messy data into reasoning-ready semantic knowledge.
Core capabilities:\n- Universal ingestion (PDF, DOCX, HTML, JSON, CSV, databases, APIs)\n- Automated entity and relationship extraction\n- Knowledge graph construction with entity resolution\n- Automated ontology generation and validation\n- GraphRAG (hybrid vector + graph retrieval, multi-hop reasoning)\n- Persistent semantic memory for AI agents\n- Conflict detection, deduplication, and provenance tracking
Project links:\nDocs: https://hawksight-ai.github.io/semantica/\nGitHub: https://github.com/Hawksight-AI/semantica
I\u2019d appreciate feedback from people working on knowledge graphs, GraphRAG, agent memory, or production RAG reliability.
Happy to discuss design trade-offs or answer technical questions."},"title":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["graphrag"],"value":"Semantica \u2013 Open-source semantic layer and GraphRAG framework"}},"_tags":["story","author_kaifahmad1","story_46452014","ask_hn"],"author":"kaifahmad1","children":[46452016],"created_at":"2026-01-01T07:13:34Z","created_at_i":1767251614,"num_comments":0,"objectID":"46452014","points":8,"story_id":46452014,"story_text":"Hi HN,
I\u2019m sharing Semantica, an MIT-licensed open-source framework for building semantic layers and knowledge engineering systems for AI.
Many RAG and agent systems fail not due to model quality, but due to the semantic gap \u2014 unstructured, inconsistent data without explicit entities, rules, or relationships. Vector-only approaches often hallucinate or fail silently under real-world data.
Semantica focuses on transforming messy data into reasoning-ready semantic knowledge.
Core capabilities:\n- Universal ingestion (PDF, DOCX, HTML, JSON, CSV, databases, APIs)\n- Automated entity and relationship extraction\n- Knowledge graph construction with entity resolution\n- Automated ontology generation and validation\n- GraphRAG (hybrid vector + graph retrieval, multi-hop reasoning)\n- Persistent semantic memory for AI agents\n- Conflict detection, deduplication, and provenance tracking
Project links:\nDocs: https://hawksight-ai.github.io/semantica/\nGitHub: https://github.com/Hawksight-AI/semantica
I\u2019d appreciate feedback from people working on knowledge graphs, GraphRAG, agent memory, or production RAG reliability.
Happy to discuss design trade-offs or answer technical questions.","title":"Semantica \u2013 Open-source semantic layer and GraphRAG framework","updated_at":"2026-03-05T23:17:12Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"sareada52"},"title":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["graphrag"],"value":"Options for Building GraphRAG: Frameworks, Graph Databases, and Tools"},"url":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["graphrag"],"value":"https://memgraph.com/blog/building-graphrag-frameworks-tools-databases"}},"_tags":["story","author_sareada52","story_43888597"],"author":"sareada52","children":[43888650],"created_at":"2025-05-04T18:49:33Z","created_at_i":1746384573,"num_comments":0,"objectID":"43888597","points":8,"story_id":43888597,"title":"Options for Building GraphRAG: Frameworks, Graph Databases, and Tools","updated_at":"2025-05-05T02:42:53Z","url":"https://memgraph.com/blog/building-graphrag-frameworks-tools-databases"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"BerislavLopac"},"title":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["graphrag"],"value":"Beyond LLMs: Building a Graph-RAG Agentic Architecture for Faster ECM Automation"},"url":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["graphrag"],"value":"https://medium.com/@hellorahulk/beyond-llms-building-a-graph-rag-agentic-architecture-for-70-faster-ecm-automation-299b05d026fb"}},"_tags":["story","author_BerislavLopac","story_45972475"],"author":"BerislavLopac","created_at":"2025-11-18T21:36:42Z","created_at_i":1763501802,"num_comments":0,"objectID":"45972475","points":7,"story_id":45972475,"title":"Beyond LLMs: Building a Graph-RAG Agentic Architecture for Faster ECM Automation","updated_at":"2026-03-05T23:02:28Z","url":"https://medium.com/@hellorahulk/beyond-llms-building-a-graph-rag-agentic-architecture-for-70-faster-ecm-automation-299b05d026fb"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"vasa_"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["graphrag"],"value":"Hey there HN! We\u2019re Vasilije, Boris, and Laszlo, and we\u2019re excited to introduce cognee, an open-source Python library that approaches building evolving semantic memory using knowledge graphs + data pipelines
Before we built cognee, Vasilije(B Economics and Clinical Psychology) worked at a few unicorns (Omio, Zalando, Taxfix), while Boris managed large-scale applications in production at Pera and StuDocu. Laszlo joined after getting his PhD in Graph Theory at the University of Szeged.
Using LLMs to connect to large datasets (RAG) has been popularized and has shown great promise. Unfortunately, this approach doesn\u2019t live up to the hype.
Let\u2019s assume we want to load a large repository from GitHub to a vector store.\nConnectingfiles in larger systems with RAG would fail because a fixed RAG limit is too constraining in longer dependency chains. While we need results that are aware of the context of the whole repository, RAG\u2019s similarity-based retrieval does not capture the full context of interdependent files spread across the repository.
This approach allows cognee to retrieve all relevant and correct context at inference time. For example, if `function A` in one file calls `function B` in another file, which calls `function C` in a third file, all code and summaries that further explain their position and purpose in that chain are served as context. As a result, the system has complete visibility into how different code parts work together within the repo.
Last year, Microsoft took a leap published GraphRAG - i.e. RAG with Knowledge Graphs. We think it is the right direction.\nOur initial ideas were similar to this paper and this got some attention on Twitter (https://x.com/tricalt/status/1722216426709365024)
Over time we understood we needed tooling to create dynamically evolving groups of graphs, cross-connected and evaluated together.\nOur tool is named after a process called cognification. We prefer the definition that Vakalo (1978) uses to explain that cognify represents "building a fitting (mental) picture"
We believe that agents of tomorrow will require a correct dynamic \u201cmental picture\u201d or context to operate in a rapidly evolving landscape.
To address this, we built ECL pipelines, where we do the following:\n- Extract data from various sources using dlt and existing frameworks\n- Cognify - create a graph/vector representation of the data\n- Load - store the data in the vector (in this case our partner FalkorDB), graph, and relational stores
We can also continuously feed the graph with new information, and when testing this approach we found that on HotpotQA, with human labeling, we achieved 87% answer accuracy (https://docs.cognee.ai/evaluations).
To show how the approach works we did an integration with continue.dev and built a codegraph
Here is how codegraph was implemented: \nWe're explicitly including repository structure details and integrating custom dependency graph versions. Think of it as a more insightful way to understand your codebase's architecture.\nBy transforming dependency graphs into knowledge graphs, we're creating a quick, graph-based version of tools like tree-sitter. This means faster and more accurate code analysis.\nWe worked on modeling causal relationships within code and enriching them with LLMs. This helps you understand how different parts of your code influence each other.\nWe created graph skeletons in memory which allows us to perform various operations on graphs and power custom retrievers.
If you want to integrate cognee into your systems or have a look at codegraph, our GitHub repository is (https://github.com/topoteretes/cognee)
Thank you for reading! We\u2019re definitely early and welcome your ideas and experiences as it relates to agents, graphs, evals, and human+LLM memory."},"title":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["graphrag"],"value":"Show HN: Cognee \u2013 Turn RAG and GraphRAG into custom dynamic semantic memory"},"url":{"matchLevel":"none","matchedWords":[],"value":"https://github.com/topoteretes/cognee"}},"_tags":["story","author_vasa_","story_43031915","show_hn"],"author":"vasa_","children":[43034688],"created_at":"2025-02-13T01:59:26Z","created_at_i":1739411966,"num_comments":1,"objectID":"43031915","points":6,"story_id":43031915,"story_text":"Hey there HN! We\u2019re Vasilije, Boris, and Laszlo, and we\u2019re excited to introduce cognee, an open-source Python library that approaches building evolving semantic memory using knowledge graphs + data pipelines
Before we built cognee, Vasilije(B Economics and Clinical Psychology) worked at a few unicorns (Omio, Zalando, Taxfix), while Boris managed large-scale applications in production at Pera and StuDocu. Laszlo joined after getting his PhD in Graph Theory at the University of Szeged.
Using LLMs to connect to large datasets (RAG) has been popularized and has shown great promise. Unfortunately, this approach doesn\u2019t live up to the hype.
Let\u2019s assume we want to load a large repository from GitHub to a vector store.\nConnectingfiles in larger systems with RAG would fail because a fixed RAG limit is too constraining in longer dependency chains. While we need results that are aware of the context of the whole repository, RAG\u2019s similarity-based retrieval does not capture the full context of interdependent files spread across the repository.
This approach allows cognee to retrieve all relevant and correct context at inference time. For example, if `function A` in one file calls `function B` in another file, which calls `function C` in a third file, all code and summaries that further explain their position and purpose in that chain are served as context. As a result, the system has complete visibility into how different code parts work together within the repo.
Last year, Microsoft took a leap published GraphRAG - i.e. RAG with Knowledge Graphs. We think it is the right direction.\nOur initial ideas were similar to this paper and this got some attention on Twitter (https://x.com/tricalt/status/1722216426709365024)
Over time we understood we needed tooling to create dynamically evolving groups of graphs, cross-connected and evaluated together.\nOur tool is named after a process called cognification. We prefer the definition that Vakalo (1978) uses to explain that cognify represents "building a fitting (mental) picture"
We believe that agents of tomorrow will require a correct dynamic \u201cmental picture\u201d or context to operate in a rapidly evolving landscape.
To address this, we built ECL pipelines, where we do the following:\n- Extract data from various sources using dlt and existing frameworks\n- Cognify - create a graph/vector representation of the data\n- Load - store the data in the vector (in this case our partner FalkorDB), graph, and relational stores
We can also continuously feed the graph with new information, and when testing this approach we found that on HotpotQA, with human labeling, we achieved 87% answer accuracy (https://docs.cognee.ai/evaluations).
To show how the approach works we did an integration with continue.dev and built a codegraph
Here is how codegraph was implemented: \nWe're explicitly including repository structure details and integrating custom dependency graph versions. Think of it as a more insightful way to understand your codebase's architecture.\nBy transforming dependency graphs into knowledge graphs, we're creating a quick, graph-based version of tools like tree-sitter. This means faster and more accurate code analysis.\nWe worked on modeling causal relationships within code and enriching them with LLMs. This helps you understand how different parts of your code influence each other.\nWe created graph skeletons in memory which allows us to perform various operations on graphs and power custom retrievers.
If you want to integrate cognee into your systems or have a look at codegraph, our GitHub repository is (https://github.com/topoteretes/cognee)
Thank you for reading! We\u2019re definitely early and welcome your ideas and experiences as it relates to agents, graphs, evals, and human+LLM memory.","title":"Show HN: Cognee \u2013 Turn RAG and GraphRAG into custom dynamic semantic memory","updated_at":"2026-06-26T19:29:14Z","url":"https://github.com/topoteretes/cognee"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"laminarflow027"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["graphrag"],"value":"In this post, we document the results of some experiments comparing vanilla Graph RAG (just a single pass of text2cypher) vs. a router agent Graph RAG approach that can call vector search tools alongside text2cypher. The routing agent uses an LLM to decide which vector search tool to call, depending on the terms identified in the question, and it works quite well.
The results show that recent frontier LLMs like `gpt-4.1` and the trusty workhorse `gemini-2.0-flash` produce great quality Cypher reliably and reproducibly, with some prompt engineering to ensure that the graph schema is formatted well in the text2cypher prompt. Across a suite of 10 test queries (that are moderately complex and require paths to be retrieved from the knowledge graph), `gpt-4.1` and `gemini-2.0-flash` pass all tests, generating the right answers when a router agent is added to the workflow to enhance vanilla Graph RAG.
Prompt engineering is done using BAML (a programming language that makes it simple to prompt LLMs and get structured outputs from them in all experiments. In fact, the knowledge graph itself was constructed using BAML prompts that extract entities and relationships from unstructured data upstream.
A logical next step to this workflow is to build more complex agent loops that can run multi-step Cypher queries whose results can be consolidated to answer harder questions (similar to how a human would approach it). The general principles of testing and evaluation would apply here, too. Seems promising to explore these methods further!"},"title":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["graphrag"],"value":"Show HN: Towards agentic Graph RAG: Enhancing graph retrieval with vector search"},"url":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["graphrag"],"value":"https://blog.kuzudb.com/post/enhancing-graph-rag-with-vector-search/"}},"_tags":["story","author_laminarflow027","story_44381939","show_hn"],"author":"laminarflow027","created_at":"2025-06-25T21:21:08Z","created_at_i":1750886468,"num_comments":0,"objectID":"44381939","points":6,"story_id":44381939,"story_text":"In this post, we document the results of some experiments comparing vanilla Graph RAG (just a single pass of text2cypher) vs. a router agent Graph RAG approach that can call vector search tools alongside text2cypher. The routing agent uses an LLM to decide which vector search tool to call, depending on the terms identified in the question, and it works quite well.
The results show that recent frontier LLMs like `gpt-4.1` and the trusty workhorse `gemini-2.0-flash` produce great quality Cypher reliably and reproducibly, with some prompt engineering to ensure that the graph schema is formatted well in the text2cypher prompt. Across a suite of 10 test queries (that are moderately complex and require paths to be retrieved from the knowledge graph), `gpt-4.1` and `gemini-2.0-flash` pass all tests, generating the right answers when a router agent is added to the workflow to enhance vanilla Graph RAG.
Prompt engineering is done using BAML (a programming language that makes it simple to prompt LLMs and get structured outputs from them in all experiments. In fact, the knowledge graph itself was constructed using BAML prompts that extract entities and relationships from unstructured data upstream.
A logical next step to this workflow is to build more complex agent loops that can run multi-step Cypher queries whose results can be consolidated to answer harder questions (similar to how a human would approach it). The general principles of testing and evaluation would apply here, too. Seems promising to explore these methods further!","title":"Show HN: Towards agentic Graph RAG: Enhancing graph retrieval with vector search","updated_at":"2025-06-27T06:37:16Z","url":"https://blog.kuzudb.com/post/enhancing-graph-rag-with-vector-search/"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"dmezzetti"},"title":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["graphrag"],"value":"GraphRAG with Wikipedia"},"url":{"matchLevel":"none","matchedWords":[],"value":"https://neuml.hashnode.dev/advanced-rag-with-graph-path-traversal"}},"_tags":["story","author_dmezzetti","story_40873973"],"author":"dmezzetti","created_at":"2024-07-04T11:17:10Z","created_at_i":1720091830,"num_comments":0,"objectID":"40873973","points":6,"story_id":40873973,"title":"GraphRAG with Wikipedia","updated_at":"2024-09-20T17:25:38Z","url":"https://neuml.hashnode.dev/advanced-rag-with-graph-path-traversal"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"daleverett"},"title":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["graphrag"],"value":"Show HN: Instant GraphRAG over any Postgres database"},"url":{"matchLevel":"none","matchedWords":[],"value":"https://polygres.com"}},"_tags":["story","author_daleverett","story_48824102","show_hn"],"author":"daleverett","children":[48824150,48824465,48834047],"created_at":"2026-07-07T21:28:59Z","created_at_i":1783459739,"num_comments":4,"objectID":"48824102","points":5,"story_id":48824102,"title":"Show HN: Instant GraphRAG over any Postgres database","updated_at":"2026-07-08T18:14:42Z","url":"https://polygres.com"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"chandanrajah"},"title":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["graphrag"],"value":"Show HN: Post-graph-RAG \u2013 your RAG still thinks the old CFO is the CFO"},"url":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["graphrag"],"value":"https://github.com/crajah/post-graph-rag"}},"_tags":["story","author_chandanrajah","story_49645202","show_hn"],"author":"chandanrajah","children":[49645207,49645216,49645221,49645226,49645230,49645589],"created_at":"2026-09-10T15:15:54Z","created_at_i":1789053354,"num_comments":0,"objectID":"49645202","points":5,"story_id":49645202,"title":"Show HN: Post-graph-RAG \u2013 your RAG still thinks the old CFO is the CFO","updated_at":"2026-09-11T17:13:22Z","url":"https://github.com/crajah/post-graph-rag"}],"hitsPerPage":20,"nbHits":962,"nbPages":49,"page":0,"params":"query=GraphRAG&tags=story&advancedSyntax=true&analyticsTags=backend","processingTimeMS":8,"processingTimingsMS":{"_request":{"roundTrip":15},"afterFetch":{"format":{"total":1}},"fetch":{"query":6,"total":7},"total":8},"query":"GraphRAG","serverTimeMS":9}