Graphiti is Knowledge Graph-based memory for AI agents. Automatically build rich graphs from changing business data & chat histories. Enable your Python agent with fast access to relevant, accurate data, even as it evolves over time. Visit our GitHub repo!
I want to introduce you to Graphiti, an open-source Temporal Knowledge Graph framework that gives AI agents the ability to learn and retain information over time, just like humans do. 🤖
Graphiti was inspired by Microsoft GraphRAG, but has a key architectural difference: It understands how newly ingested data might change existing data, and is purpose built for data that evolves over time.
Why does this matter? Imagine you tell a food delivery app that you've adopted veganism, but it keeps recommending your once-favorite burger joint. Or picture a sales assistant that forgets a key client's purchasing history and constraints, resulting in tone-deaf product pitches. That's the kind of experience that frustrates users and undermines trust in AI.
With Graphiti, AI agents can reason with evolving data, enabling more personalized, context-aware interactions. Whether you're building a chatbot, LLM-powered assistant, or a next-gen AI system, Graphiti provides the foundation for fast, accuratelong-term memory.
We've implemented Graphiti in Zep, our memory layer for enterprise AI agents, and wrote a paper on how Zep and Graphiti perform versus other approaches to agent memory (it's the State of the Art!). You can find an overview here or read the paper on arXiv.
Looks really interesting! Curious to see how this can be implemented for the whole team on GitHub.
Love the focus on memory! AI that understands changes in data instead of just storing static information is a major step forward. What industries do you see benefiting from this the most?
Awesome product @daniel_chalef and Team! Daniel you're mentioning enterprise AI use cases for Graphiti. Indeed there are legions of marketing managers monitoring monthly KPIs across multiple product lines based on guts feeling who would benefit from using it for analytics. Best of luck!
Congrats on the launch! We continue to see across @Langfuse users that memory is crucial for creating meaningful agents capable of engaging in long conversations and providing a personalized experience. Well done, Team Zep!
I'm genuinely impressed. @Graphiti might solve the long lived issue we've just discussed with the team in application to agentical memory being "not human enough".
This is a huge step forward for AI memory and personalization. The problem of AI forgetting or misinterpreting evolving user preferences is one of the biggest blockers to truly intelligent assistants. Graphiti’s approach to dynamically updating knowledge instead of just appending new data is huge.
Congrats on the launch!
Best wishes and sending lots of wins to the team :) @daniel_chalef
Graphiti’s approach to evolving data feels like a breakthrough, especially for systems where user context shifts over time. It’s refreshing to see a framework that keeps AI “memory” accurate and current. Can’t wait to explore it for context-heavy chatbots!
Huge potential here! AI-powered customer support, sales assistants, and chatbots could all benefit from better long-term memory. Wishing you all the success with Graphiti!
Excited to see this in action! AI remembering past interactions is key for personalization—Graphiti sounds like the missing piece. What’s next on your roadmap?
I’m excited about the potential for community collaboration. I also appreciate how Graphiti can ingest structured and unstructured data seamlessly. It’s amazing for a variety of projects.
Does it handle forgetting outdated info automatically or do I need to set explicit rules?
Absolutely! Lately, I’ve been feeling drained from repeatedly reminding chatbots to remember the same things. This is truly impressive! Wishing you great success! 🚀
Graphiti gives chatbots memory that updates over time. 🧠⌛
Graphiti's approach is superior to classic RAG because it stores information in a queryable Knowledge Graph which replaces big text dumps that usually overstuff context windows.
Not only does this mean you no longer have to keep reminding chatbots of things you've already told them, but they can start to reason over data and information that changes over time, like your affinity for certain brands or the number of pets or relatives you might need to take on a trip.
