Artificial Intelligence

Graphon Labs Raises $35M to Build Persistent Memory for AI Agents

By mohitdeep1122 • • 6 min read • 📷 3 High-Res Visuals
Share Dispatch
𝕏 Post in Share Y Hacker News
⚡ TL;DR — Executive Summary

Graphon Labs raises $35M in Series A funding led by RRE Ventures to build persistent AI memory connecting documents, videos, images and enterprise data.

Graphon Labs raises $35 million in Series A funding to develop persistent AI intelligence infrastructure that helps agents retain and connect information over time.
Graphon Labs raises $35 million in Series A funding to develop persistent AI intelligence infrastructure that helps agents retain and connect information over time. 🔍 Click to zoom

AI agents can write code, analyse documents and execute complex tasks, but they still face a fundamental limitation: remembering everything they have learned and understanding how that information connects over time. A conversation from last month, an engineering video from last year and a customer document stored in another system may all contain information needed to solve one problem. San Francisco-based Graphon Labs is building infrastructure to connect these disconnected sources into a persistent intelligence layer for AI. The company has raised $35 million in a Series A round led by RRE Ventures, with participation from Energy Impact Partners, Hitachi Ventures, Gaia VC and HP. The investment brings its reported total funding to $43.3 million and will accelerate development of a memory platform designed for enterprise AI agents and developers.

Graphon Labs uses graph-based intelligence to connect information across documents, images, audio and videos, helping AI systems understand relationships between different data sources.
Graphon Labs uses graph-based intelligence to connect information across documents, images, audio and videos, helping AI systems understand relationships between different data sources.🔍 Click image to enlarge

The problem with AI memory goes beyond context windows

Most AI applications depend on retrieval-augmented generation, or RAG, to bring relevant information into a model’s context window. This approach generally works by breaking documents into smaller sections, searching for relevant sections and passing the results to an AI model. However, the process can lose important relationships between different pieces of information. A maintenance instruction might be stored in a PDF, the actual machine condition might appear in a video, and a warning about the same equipment might exist in an internal report. Finding three separate pieces does not automatically explain how they connect or which information came first. Graphon argues that businesses need more than a faster document search system. They need infrastructure that can preserve relationships, history, sources and context across their entire information environment.

Graphon uses mathematical graphons to connect fragmented information

The company’s technology is built around graphons, a mathematical framework for representing the structure of very large networks. Graphon Labs uses this approach to create a detailed graph of information and relationships across different data sources, then builds a compressed mathematical representation to guide searches across that larger structure. Its technology is designed to process documents, images, audio and videos, automatically identifying information and relationships without requiring companies to manually define every connection in advance. For example, an industrial AI assistant could connect a machine’s maintenance history, a technician’s video explanation and an engineering document instead of treating them as unrelated files. Graphon says this structure allows AI systems to retrieve more complete evidence and understand how separate events and facts relate to one another.

It wants to become infrastructure that any AI agent can use

Graphon is not trying to replace the foundation models built by OpenAI, Anthropic or Google. Instead, it wants to provide the information layer those models draw upon. Developers can connect existing agents, including Claude and GPT-based applications, to Graphon through APIs and its Model Context Protocol server. Its developer platform combines S3-compatible object storage, keyword and semantic search, source-cited question answering and deeper research across large collections of data. The platform also supports applications such as enterprise knowledge bases, video libraries, internal search and long-term agent memory. This makes the business opportunity broader than conversational memory alone: Graphon wants to become a shared knowledge infrastructure that software companies can use to give their AI applications access to information that remains available beyond individual conversations.

Graphon's persistent memory platform aims to give enterprise AI agents access to long-term knowledge, contextual information and connected business data.
Graphon's persistent memory platform aims to give enterprise AI agents access to long-term knowledge, contextual information and connected business data.🔍 Click image to enlarge

Enterprise interest is already emerging across different industries

Graphon is targeting organisations that manage large quantities of technical, operational and multimedia information, particularly in manufacturing, financial services, healthcare, media and sports. Business Insider reported that the company already has software users including HP and the San Francisco 49ers. The Next Web also reported that the 49ers were exploring applications such as identifying sponsorship opportunities, while Hitachi Rail described its evaluation of the platform as showing promise in an industrial setting. These examples illustrate why persistent multimodal memory could matter beyond conventional office documents. In manufacturing, an AI system may need to connect maintenance videos with equipment records; in financial services, it may need to connect reports, transaction history and research documents. The challenge is turning those disconnected information sources into a reliable context that employees and AI agents can actually use.

Benchmark results provide an early indication of technical progress

Graphon is also using public AI memory benchmarks to demonstrate its approach. The company reports approximately 94% accuracy on LoCoMo, a benchmark that tests an AI system’s ability to remember and reason across lengthy conversations, with about 7,900 tokens used per question. It also reports an 81.7% question-answering score on ATM-Bench, which evaluates memory and retrieval across images, videos, audio and text. The public ATM-Bench leaderboard lists a Graphon V1 run at 81.71% on its question-answering score and 89.78% on Recall@10. These are meaningful early technical signals, but benchmark performance should not be confused with proven reliability across every enterprise environment. Graphon’s own reported results on the harder ATM-Bench track were still awaiting leaderboard review according to its benchmark disclosures. Real-world performance will ultimately depend on data quality, retrieval accuracy, operating costs and how well the system handles changing information.

Funding will support research, enterprise deployment and wider access

Graphon Labs was founded in July 2025 by Arbaaz Khan, Clark Zhang and Deepak Mishra, whose research backgrounds include work at the University of Pennsylvania’s GRASP Lab on graph-based systems for robotics and memory. The startup previously raised $8.3 million in seed funding in May 2026, led by Novera Ventures, with participation from investors including Perplexity Fund, Samsung Next and Hitachi Ventures. Its new $35 million financing will primarily support engineering and research hiring, alongside sales and go-to-market expansion. Graphon currently offers deployment through its cloud, a customer’s own virtual private cloud and on-device environments, while also reporting SOC 2 Type II certification. The broader ambition is to make its intelligence infrastructure available across enterprise systems, developer applications and eventually personal devices.

Persistent intelligence could become another essential layer of AI infrastructure

The long-term opportunity for Graphon is tied to a major shift in how artificial intelligence is being deployed. As companies move from asking AI models individual questions towards building agents that continuously perform work, the ability to preserve information and reason across different sources becomes increasingly important. However, Graphon is entering a competitive market where memory platforms, database companies, search providers and foundation-model developers are all working to improve long-term context. Its challenge will be to prove that its graphon-based architecture offers meaningful advantages over existing retrieval and knowledge systems in accuracy, cost, speed and ease of deployment. If it succeeds, Graphon could become an important layer between enterprise data and the AI systems that use it. Its $35 million Series A gives the company the resources to pursue that ambition; the next milestone is demonstrating that persistent memory can work reliably at commercial scale.

Sources and further reading

📷 Complete Article Visual Gallery (3 Pictures) Click any photo for full-screen view
Graphon Labs raises $35 million in Series A funding to develop persistent AI intelligence infrastructure that helps agents retain and connect information over time.
Graphon Labs raises $35 million in Series A funding to develop persistent AI intelligence infrastructure that helps agents retain and connect information over time.
Graphon Labs uses graph-based intelligence to connect information across documents, images, audio and videos, helping AI systems understand relationships between different data sources.
Graphon Labs uses graph-based intelligence to connect information across documents, images, audio and videos, helping AI systems understand relationships between different data sources.
Graphon's persistent memory platform aims to give enterprise AI agents access to long-term knowledge, contextual information and connected business data.
Graphon's persistent memory platform aims to give enterprise AI agents access to long-term knowledge, contextual information and connected business data.
mohitdeep1122 • TechFront360 Editorial Desk

Published by TechFront360.com following our strict fact-checking, benchmark verification, and independent technology journalism standards.

More in Artificial Intelligence