Graphon Labs raises $35M Series A to give AI agents persistent memory

San Francisco-based Graphon Labs announced on November 10, 2026 a $35 million Series A led by RRE Ventures, with Energy Impact Partners, Hitachi Ventures, Gaia VC and HP also participating, bringing the company's disclosed financing to $43.3 million less than six months after it left stealth with an $8.3 million seed round.

What Graphon actually builds

Graphon's platform ingests documents, video, images, audio and logs, discovers how the pieces relate to each other, and stores that as a persistent, queryable graph any model or agent can call on later. Rather than requiring a team to specify a schema up front, the system is described by Yahoo Finance as self-supervised and schema-free, built on graphons, a mathematical framework for analyzing large graphs, and already deployed with enterprises including Hitachi and the San Francisco 49ers. Pulse2's coverage of the round frames the target problem plainly: agent systems can write code and query databases well enough, but they routinely lose track of prior interactions or fail to reconcile company information that keeps changing underneath them. Graphon's pitch is a graph layer that preserves provenance and tracks how facts evolve, so an agent is reasoning against a current, traceable record rather than a stale training snapshot.

Schema-free memory versus a specified ontology

Graphon's graph is discovered from the data itself, which lets it start ingesting unstructured material immediately without a modeling phase. A mid-level ontology like CCO takes the opposite starting position: categories and relations are specified in advance, logically axiomatized, and built to extend a top-level ontology (BFO) so that two independently built systems mapped against the same suite stay interoperable by construction, not by inference after the fact.

Schema-free discovery is well suited to surfacing relationships nobody anticipated in a messy document corpus. A specified ontology is better suited to the cases Graphon's own use case ultimately depends on: making sure "customer," "shipment," or "incident" means the same committed thing everywhere an agent encounters it, including systems that were never trained together.

Why the round matters beyond Graphon

Funding rounds at this size for agent-memory infrastructure signal that enterprises are already past the pilot-chatbot stage and are now funding the plumbing underneath multi-agent deployments. Standing up that plumbing against one real, bounded process first, before a schema (specified or discovered) gets extended across an entire organization, is usually where an outside team doing focused custom software development adds the most value: building the narrow data-integration layer that proves an agent's answers actually hold up against a company's real records, before the bigger rollout gets committed to.

What to watch

Graphon says the new capital goes toward expanding enterprise deployments and opening the infrastructure to more developers. Whether a self-supervised, schema-free graph stays internally consistent as dozens of independently built agents write into and read from it at enterprise scale is the detail worth watching as Graphon's enterprise footprint grows past its initial anchor deployments. That kind of drift is exactly the failure mode a specified, axiomatized ontology is designed to prevent.