Graphon AI raises $35M to fix AI agents’ memory problem with graph-based relational storag

San Francisco-based Graphon AI has raised $35 million in Series A funding, led by RRE Ventures with participation from HP and Hitachi Ventures. The round arrives just six months after its $8.3 million seed, a rapid pace for a company founded in July 2025.

The core problem Graphon is solving is simple: AI agents today forget. They rely on fixed context windows—limited chunks of text they can process at once—meaning any information outside that window is lost. Graphon’s solution is a “pre-model intelligence layer” that ingests unstructured data (documents, videos, images, logs) and builds a self-supervised knowledge graph, mapping relationships without requiring human-defined schemas. Instead of treating data as isolated fragments, the system preserves how pieces connect and evolve over time.

The technology leverages graphon functions—a mathematical framework for modeling relationships in large networks—to create a dynamic, time-aware knowledge base. Graphon claims strong benchmark results, including first place on ATM-Bench, 94% on LoCoMo, and 84% on DolphinBench. More importantly, its model-agnostic design means it can integrate with existing models like GPT or Claude, avoiding vendor lock-in—a key concern for enterprises evaluating AI tools.

The investor lineup suggests a clear target market: large corporations. Backers like HP and Hitachi Ventures, along with an early deployment with South Korea’s GS Group, point to enterprise adoption. The inclusion of the San Francisco 49ers as a partner adds an interesting data point—if Graphon can solve problems for a conglomerate, a hardware maker, and a sports organization, it may indicate broad applicability.

The timing is significant. While AI agents are proliferating—from coding assistants to IT operations—most still rely on workarounds like retrieval-augmented generation (RAG), which fetches disconnected data chunks when needed. Graphon’s alternative is a knowledge graph that maintains relationships dynamically, a potential advantage for use cases where context matters, such as compliance, customer service, or long-term decision-making.

That said, questions remain. The company is still young, and its benchmarks don’t yet prove real-world scalability. Competitors like WhiteMatter (which raised $20 million this year for AI agent governance) or Uvi (which secured $7.5 million to cut agent costs) are addressing adjacent challenges. Whether enterprises will adopt Graphon’s memory layer—or opt for simpler, modular solutions—isn’t yet clear.

One thing is certain: Graphon is staking a claim in a niche but growing segment of AI infrastructure. As agents become more sophisticated, the need for persistent memory will only grow. The next steps will be watching how its early partners deploy the technology and whether it can move beyond proof-of-concept use cases.

Sources: cryptobriefing.com