Graphon Labs
Provides persistent memory for AI agents by mapping relationships across multimodal enterprise data.
Website: https://www.graphonlabs.ai
Cover Block
Open sources
| Field | Value |
|---|---|
| Name | Graphon Labs |
| Tagline | Provides persistent memory for AI agents by mapping relationships across multimodal enterprise data [LinkedIn] |
| Headquarters | San Francisco, US [LinkedIn] |
| Founded | 2025 [LinkedIn] |
| Stage | Series A [The Next Web, November 2026] |
| Business Model | SaaS |
| Industry | Deeptech |
| Technology | AI / Machine Learning |
| Geography | North America |
| Growth Profile | Venture Scale |
| Founding Team | Co-Founders (5): Arbaaz Khan, Clark Zhang, Deepak Mishra, Siddharth Mysore, Jing Li [Business Insider, November 2026] [Siddharth Mysore LinkedIn, Retrieved 2026] [Jing Li LinkedIn, Retrieved 2026] |
| Funding Label | $10M+ |
| Total Disclosed Funding | $43.3 million as of November 2026 [Rutland Herald, November 2026] |
Links
Open sources
What an Investor Needs First
PUBLIC Graphon Labs is building a memory layer for AI agents, and it merits investor attention now because it moved from an $8.3 million seed in May 2026 to a $35 million Series A in November 2026, an unusually fast financing cadence for an infrastructure company still defining a new category [AI Journal, May 2026] [The Next Web, November 2026] [Rutland Herald, November 2026]. The company, based in San Francisco and founded in 2025, frames its system as persistent memory that maps relationships across enterprise documents, video, audio, images, and operational data so software agents can retain context over time rather than retrieving isolated snippets at query time [LinkedIn] [Business Insider, November 2026] [Siddharth Mysore LinkedIn].
The founding story is unusually academic for a young applied AI company: coverage ties the name "Graphon" to CEO Arbaaz Khan's Ph.D. work on graphons in robot-swarm research, and public materials identify Khan, Clark Zhang, and Deepak Mishra as co-founders, with Siddharth Mysore and Jing Li also associated with the company [Business Insider, November 2026] [Sohail Khalid LinkedIn] [Jing Li LinkedIn]. That matters because the product pitch rests less on a new foundation model and more on a graph-based pre-model intelligence layer, a technical position that could resonate with enterprises that want retrieval, reasoning, and governance improvements without rebuilding their agent stack from scratch [AI Journal, May 2026] [Penn Engineering / GRASP, 2026] [The Next Web, November 2026].
Public reporting describes a three-part product surface: Graphon Enterprise for governed enterprise deployments, Graphon Developer for builders working across data types, and an on-device memory product aimed at endpoint use cases; sources also report compatibility with off-the-shelf agents such as Claude and GPT [The Next Web, November 2026] [Business Insider, November 2026]. The business model is presented as SaaS, but the more material point for investors is whether this becomes a system-of-record layer inside large accounts or remains a useful add-on for agent experiments [LinkedIn] [The Next Web, November 2026].
The cap table is notable for strategic density. Seed investors included Novera Ventures, Perplexity Fund, Samsung Next, GS Futures, Hitachi Ventures, Gaia Ventures, B37 Ventures, and Aurum Partners, while the Series A was led by RRE Ventures with participation from HP, Energy Impact Partners, Hitachi Ventures, and Gaia VC, bringing total disclosed funding to $43.3 million as of November 2026 [AI Journal, May 2026] [The Next Web, November 2026] [Rutland Herald, November 2026]. Over the next 12 to 18 months, the key question is not whether the memory problem exists, but whether Graphon can convert early credibility with reported users such as GS Group, HP, Hitachi, and the San Francisco 49ers into repeatable enterprise deployments with clear proof that its graph layer improves agent performance, retention, or cost in production settings [AI Journal, May 2026] [Penn Engineering / GRASP, 2026] [The Next Web, November 2026].
Claim stands unchecked -- This section combines independently reported financing with several company-linked or profile-based product and team claims, and key product details remain only partially corroborated by public independent sources.
Taxonomy Snapshot
| Axis | Value |
|---|---|
| Stage | Series A |
| Business Model | SaaS |
| Industry / Vertical | Deeptech |
| Technology Type | AI / Machine Learning |
| Geography | North America |
| Growth Profile | Venture Scale |
| Founding Team | Co-Founders (3+) |
| Funding | $43.3M total disclosed |
Inside the Company
PUBLIC Graphon Labs appears to be a very young San Francisco company organized around a specific infrastructure problem in AI, persistent memory for agents, rather than a general model-development pitch. Public materials identify the company as based in San Francisco and founded in 2025, with a LinkedIn company page describing an AI-focused organization under the Graphon Labs name [LinkedIn]. The available public record in this section does not establish a legal entity filing with comparable clarity, so the operating-company description is best kept at the level of headquarters, founding year, and stated focus [LinkedIn].
The chronology is short but unusually compressed. Graphon emerged from stealth alongside an $8.3 million seed round in May 2026, according to coverage that cited the company announcement, and by November 2026 it had announced a $35 million Series A, bringing total disclosed funding to $43.3 million [AI Journal, May 2026] [Rutland Herald, November 2026]. LinkedIn also supports that the company was operating by 2026 under the Graphon Labs name from its San Francisco base, which is directionally consistent with the funding timeline even if it does not independently verify every milestone date [LinkedIn].
The founding story is only partly documented in sources that meet the citation constraint for this section. Public reporting outside this section states that the name draws from co-founder Arbaaz Khan's academic work on graphons, but that attribution is not corroborated here by a company website, Crunchbase, or state filing, so it should be treated as unconfirmed within the overview itself [LinkedIn]. What is clear from the narrow source set is that Graphon was founded recently, is headquartered in San Francisco, and moved from founding to seed and then Series A financing within roughly one year of the stated 2025 founding date [LinkedIn] [AI Journal, May 2026] [Rutland Herald, November 2026].
Partially corroborated -- This section relies primarily on LinkedIn, with milestone support from named funding coverage rather than company website, Crunchbase, or state filings.
Under the Hood
MIXED
The product claim here is straightforward, even if the evidence base is still thin. Graphon Labs says it provides persistent memory for AI agents by mapping relationships across multimodal enterprise data, including documents, video, audio, images, and operational systems, so agents can retain context and retrieve information over time [Business Insider, November 2026] [The Next Web, November 2026]. In founder-linked descriptions, the company frames this as a pre-model intelligence layer that organizes raw unstructured data before a model answers a query, rather than relying only on retrieval of isolated snippets [AI Journal, May 2026] [LinkedIn].
Public reporting also points to a three-part product surface, though much of the feature detail still traces back to company descriptions repeated in coverage. Graphon Enterprise is described as a governed memory layer for enterprise data with deployment options spanning company-hosted environments, customer-managed virtual private cloud, and on-premises setups [The Next Web, November 2026] [Rutland Herald, November 2026]. Graphon Developer is positioned for builders working across multiple data types, while Graphon On-Device Memory is described as targeting phones, laptops, cameras, wearables, and related edge devices [The Next Web, November 2026] [Business Insider, November 2026]. Reporting also says the system can connect to off-the-shelf agents such as Claude and GPT, which suggests Graphon is trying to sit between enterprise data and existing model endpoints rather than competing at the foundation-model layer [Business Insider, November 2026].
What is verifiable in public is the architectural intent, not yet the operating proof. Named coverage and a university profile consistently describe the core idea as a graph-based memory layer for enterprise AI, but they do not establish benchmark results, production-scale performance, or detailed integration evidence across the claimed modalities and deployment modes [Penn Engineering / GRASP] [The Next Web, November 2026] [Business Insider, November 2026]. That leaves the technology concept legible, and potentially differentiated if it works as described, while keeping the current read squarely at the product-definition stage rather than the fully validated platform stage.
Partially corroborated -- Core product framing is corroborated by multiple public outlets, but several feature and deployment details remain company-described through press coverage and founder-linked materials.
Market Research
PUBLIC
The market matters now because enterprise interest has shifted from standalone large language models to the harder problem of giving agents durable context across messy internal data, and Graphon is explicitly positioned at that layer between raw information and model inference [Business Insider, November 2026] [Penn Engineering / GRASP, 2026].
Public sources do not provide a named third-party TAM for Graphon's exact category, so the cleaner way to frame the market is through analogous markets rather than a false precision. The available reporting places Graphon against enterprise and developer demand for systems that connect documents, video, audio, images, and operational data into a persistent representation that agents can query over time [Business Insider, November 2026] [The Next Web, November 2026]. That is narrower than the full generative AI software market, but broader than classic retrieval tooling, because the company's stated wedge is memory and relationship mapping across multimodal enterprise data rather than single-query search [AI Journal, May 2026] [Penn Engineering / GRASP, 2026].
The immediate serviceable market also appears bifurcated. One side is enterprise buyers in data-heavy sectors such as industrial, financial services, healthcare, manufacturing, sports, and media, all sectors cited in the reporting as relevant targets or early usage contexts [Business Insider, November 2026] [The Next Web, November 2026]. The other is developers building agentic applications across multiple data types, which matters because Graphon has described both an enterprise product and a developer-facing product in coverage of its financing [The Next Web, November 2026] [Rutland Herald, November 2026].
| Market lens | What public sources support | Relevance to Graphon |
|---|---|---|
| Analogous market: enterprise AI infrastructure | Enterprises are seeking ways to connect internal data to model workflows, especially when retrieval and reasoning quality depend on more than document search [Penn Engineering / GRASP, 2026] [Business Insider, November 2026] | Supports the case for a middleware layer focused on memory and context persistence |
| Analogous market: multimodal data management | Graphon is described as working across documents, video, audio, images, and operational systems, which pushes it beyond text-only knowledge bases [The Next Web, November 2026] [Business Insider, November 2026] | Expands the addressable problem set into operational and media-rich environments |
| Adjacent market: on-device AI memory | Coverage says the company has an on-device memory product aimed at phones, laptops, cameras, wearables, and other devices [The Next Web, November 2026] [Rutland Herald, November 2026] | Opens a second market surface if inference shifts partly to edge environments |
The useful point in the table is not market size, which public reporting does not quantify, but market shape. Graphon is not selling into a single software budget line, according to the sources, it is trying to attach itself to enterprise AI infrastructure, multimodal data plumbing, and potentially edge AI memory at once [The Next Web, November 2026] [Business Insider, November 2026].
Demand drivers in the public record are fairly consistent. Multiple sources describe the core problem as agent "amnesia," or the inability of AI systems to retain context and reason over changing enterprise information across time [Business Insider, November 2026] [The Next Web, November 2026]. Penn Engineering's coverage also frames the product as a way to lighten the load on AI models by inserting an intelligence layer before model queries, which implies a cost and performance tailwind if enterprises decide that better structure upstream can reduce repeated context stuffing downstream [Penn Engineering / GRASP, 2026].
There is also early evidence that buyers in operationally complex organizations are testing this thesis. AI Journal and Penn Engineering both point to GS Group's 52g initiative as an early deployment context, while later reporting identifies HP, Hitachi, and the San Francisco 49ers as customers or users [AI Journal, May 2026] [Penn Engineering / GRASP, 2026] [The Next Web, November 2026]. That does not establish category maturity, contract scale, or repeatability, but it does suggest the problem is resonating in environments where information sits across many systems and modalities.
Adjacent and substitute markets matter here because Graphon's success is partly a question of what enterprises choose instead. One substitute is conventional retrieval-augmented generation stacks that index documents and return snippets at query time, the approach Graphon is implicitly arguing is insufficient for persistent agent memory [Business Insider, November 2026] [AI Journal, May 2026]. Another adjacent path is enterprise data integration and knowledge graph software, since Graphon's own framing centers on mapping relationships across organizational data before a model runs a query [Siddharth Mysore LinkedIn] [Penn Engineering / GRASP, 2026].
The regulatory and macro picture is supportive in one narrow sense and unresolved in another. Supportive, because Graphon's reported deployment options include hosted, customer-managed virtual private cloud, and on-premises environments, which aligns with the procurement realities of sectors handling sensitive data [Rutland Herald, November 2026]. Unresolved, because the same sectors named in public coverage, including healthcare and financial services, tend to impose higher standards around governance, security, and data handling, and the public record here says more about architectural intent than audited compliance or production scale [Business Insider, November 2026] [The Next Web, November 2026].
On balance, the market case rests less on a neatly bounded category and more on a real workflow gap emerging as enterprises move from chatbot experiments to agent deployment. If that shift continues, the budget could come from several adjacent pools at once, but the public evidence still stops short of proving how large a dedicated "agent memory" category becomes as a standalone spend line.
Partially corroborated -- This section relies on named publisher coverage for demand drivers, target sectors, and adjacent market framing, but it does not have a cited third-party TAM report for Graphon's exact category and therefore uses analogous-market analysis grounded in public reporting.
Competition and Substitutes
MIXED Graphon Labs is positioning itself less as another application-layer agent company and more as infrastructure that sits between enterprise data and the model, which places it in competition with several different categories at once rather than a single named peer set [The Next Web, November 2026] [Business Insider, November 2026] [Penn Engineering / GRASP, 2026].
The public record supports a three-way map. First, there are incumbent data and knowledge-system vendors that already control important enterprise workflows, even if the available source set here does not name them individually. Second, there are emerging agent-memory and retrieval challengers, where Graphon is trying to differentiate through a graph-based "Persistent Intelligence" layer that maps relationships across documents, video, audio, images, and operational systems before a model query runs [Business Insider, November 2026] [Siddharth Mysore LinkedIn]. Third, there are adjacent substitutes, including direct use of foundation-model tooling and conventional retrieval architectures, because Graphon says its products can work with off-the-shelf agents such as Claude and GPT rather than replace them outright [Business Insider, November 2026] [The Next Web, November 2026].
The edge visible today is narrow but real. Graphon has raised $43.3 million across a May 2026 seed and a November 2026 Series A, with backing that includes Novera Ventures, RRE Ventures, HP, Energy Impact Partners, Hitachi Ventures, and others, which gives it enough capital and some degree of enterprise signaling in a category where credibility matters early [AI Journal, May 2026] [The Next Web, November 2026] [Rutland Herald, November 2026]. It also has reported design points that matter in enterprise procurement, including hosted, customer-managed VPC, and on-premises deployment options through Graphon Enterprise [The Next Web, November 2026]. That said, the durability of this edge is still unproven because much of the product detail and customer narrative rests on company-linked or company-amplified reporting rather than long-cycle proof points such as renewals, production scale, or ecosystem lock-in [Business Insider, November 2026] [Penn Engineering / GRASP, 2026].
The exposure is easier to see than the moat. Graphon appears strongest when a customer believes agent memory should be its own infrastructure layer, but it is more vulnerable if buyers decide acceptable performance can be achieved inside existing model, database, or workflow stacks. The company also does not appear to own a dominant distribution channel from the public materials provided here. Reported customers include HP, Hitachi, GS Group, and the San Francisco 49ers, but the sources do not establish contract depth, deployment breadth, or whether those references convert into repeatable go-to-market advantage across a broader customer base [AI Journal, May 2026] [The Next Web, November 2026] [Penn Engineering / GRASP, 2026].
Over the next 18 months, the most plausible competitive scenario is bifurcation by buyer sophistication. Graphon is the likely winner if large enterprises keep concluding that multimodal context, governance, and deployment flexibility require a dedicated memory layer outside the base model, especially in industrial and operational environments where data types are fragmented and sensitive [Business Insider, November 2026] [Penn Engineering / GRASP, 2026]. A plausible loser if the opposite happens is Graphon itself, because the available evidence does not yet show a consumer channel, a default developer standard, or a deeply entrenched enterprise platform position that would protect it if memory features are absorbed upstream by model vendors or bundled into broader enterprise stacks. That makes the company interesting, but still category-contingent.
Partially corroborated -- Core financing facts are corroborated by multiple public outlets, but competitive positioning and product differentiation rely partly on company-linked reporting and LinkedIn-based descriptions.
Opportunity
Open sources
If Graphon executes, the prize is not a feature sale into the AI stack, it is becoming a core memory and context layer that enterprise agents rely on every time they touch fragmented, multimodal internal data [The Next Web, November 2026] [Business Insider, November 2026].
The headline opportunity is straightforward: own the persistence layer for enterprise AI agents before that layer is standardized elsewhere. Public reporting consistently describes Graphon as a graph-based system that maps relationships across documents, video, audio, images, and operational systems, then exposes that connected context to models and agents over time [Business Insider, November 2026] [Penn Engineering / GRASP, 2026] [The Next Web, November 2026]. That matters because the core failure mode in enterprise agent deployments is rarely raw model quality alone, it is memory, retrieval, and reasoning across messy internal systems, which is exactly the problem Graphon says it is built to address [The Next Web, November 2026] [AI Journal, May 2026]. The outcome remains early, but the ingredients that make it reachable are visible: a seed round in May 2026, a $35 million Series A by November 2026, and reported customer relationships that include GS Group, HP, Hitachi, and the San Francisco 49ers, all unusually strong signals for a company founded in 2025 [AI Journal, May 2026] [The Next Web, November 2026] [Rutland Herald, November 2026].
The growth paths are still contingent, but they are legible from the public record.
| Scenario | What happens | Catalyst | Why it's plausible |
|---|---|---|---|
| Industrial system of record for agents | Graphon becomes the default memory layer inside large industrial and operational environments where data lives across files, sensors, media, and legacy systems | A visible enterprise rollout with investors or customers already in the orbit, including GS Group, HP, or Hitachi [AI Journal, May 2026] [The Next Web, November 2026] | The company is already framed around multimodal operational data, and early named users come from industrial and enterprise contexts rather than consumer AI experimentation [Penn Engineering / GRASP, 2026] [The Next Web, November 2026] |
| Embedded developer infrastructure | Graphon Developer gains adoption as an API or platform layer that developers plug into off-the-shelf agents such as Claude and GPT | Broader launch and tooling around the developer product, plus reference architectures that reduce integration time [Business Insider, November 2026] | Reporting says the product is compatible with existing agents and includes a dedicated developer surface, which gives the company a path beyond direct enterprise sales [Business Insider, November 2026] [The Next Web, November 2026] |
| Edge and device memory platform | Graphon On-Device Memory turns the company into infrastructure for persistent context on phones, laptops, cameras, wearables, and adjacent hardware | A hardware distribution partnership or deeper commercialization with strategic backers such as HP or Samsung Next [The Next Web, November 2026] [AI Journal, May 2026] | The on-device product is already part of the stated roadmap, and the cap table includes investors and strategic participants with device and enterprise distribution channels [AI Journal, May 2026] [The Next Web, November 2026] |
What compounding looks like here is not a social network effect, it is workflow entrenchment. If Graphon sits between enterprise data and the model layer, each successful deployment can deepen over time as more repositories, media types, and operating systems are connected into the same memory graph, making the system more useful and harder to replace inside that account [Penn Engineering / GRASP, 2026] [Business Insider, November 2026]. The company's three-surface product strategy, enterprise, developer, and on-device, also suggests a second source of compounding: improvements to the core memory layer can travel across multiple go-to-market motions rather than being trapped in one vertical product [The Next Web, November 2026]. The early funding cadence matters here too. Moving from an $8.3 million seed in May 2026 to a $35 million Series A in November 2026 indicates investors saw enough product and customer momentum to accelerate capitalization before the category settles [AI Journal, May 2026] [Rutland Herald, November 2026].
The size of the win is easiest to frame as infrastructure value rather than near-term revenue. Public materials provided for this section do not include a third-party market sizing report or a directly comparable public peer with cited market capitalization, so any valuation framing has to stay conditional. On the evidence available, the upper-end scenario is that Graphon becomes a control-point layer in enterprise AI architecture, similar in strategic position to other infrastructure companies that sit between raw data systems and application behavior, in which case a multibillion-dollar outcome is conceivable (scenario, not a forecast), especially if the company converts early strategic relationships into standard deployment patterns across large enterprises [The Next Web, November 2026] [AI Journal, May 2026] [Business Insider, November 2026]. The public record supports the existence of that path. It does not yet support precision on how large it becomes.
Partially corroborated -- Relies on named-publisher reporting for funding, customers, and product framing, but several product details remain based on company-described materials and partially corroborated coverage.
Sources
Open sources
[LinkedIn] Graphon Labs | LinkedIn | https://www.linkedin.com/company/graphon-ai
[The Next Web, November 2026] Graphon Labs raises $35m to give AI agents a memory that lasts | https://thenextweb.com/news/graphon-labs-35m-series-a-ai-agent-memory
[AI Journal, May 2026] Graphon AI Raises $8.3M to Build Pre-Model Intelligence Layer | https://aijourn.com/graphon-ai-raises-8-3m-to-build-pre-model-intelligence-layer
[Rutland Herald, November 2026] Graphon Labs Raises $35M Series A to Accelerate AGI with Persistent Intelligence | https://www.rutlandherald.com/news/business/graphon-labs-raises-35m-series-a-to-accelerate-agi-with-persistent-intelligence/article_d4ad877e-e3c6-5ffa-8683-114511fb2ee8.html
[Business Insider, November 2026] This startup says AI agents suffer from 'amnesia.' It raised $35 million to solve that with this pitch deck. | https://www.businessinsider.com/graphon-pitch-deck-funding-ai-agents-amnesia-2026-10
[Siddharth Mysore LinkedIn, Retrieved 2026] Siddharth Mysore | LinkedIn | https://www.linkedin.com/
[Sohail Khalid LinkedIn, Retrieved 2026] Sohail Khalid | LinkedIn | https://www.linkedin.com/
[Jing Li LinkedIn, Retrieved 2026] Jing Li | LinkedIn | https://www.linkedin.com/
[Penn Engineering / GRASP, 2026] Graphon Says Its ‘Intelligence Layer’ Will Lighten the Load on AI Models | https://www.grasp.upenn.edu/news/graphon-says-its-intelligence-layer-will-lighten-the-load-on/
Articles about Graphon Labs
- Graphon Labs Is Giving AI Agents a Memory That Lasts — The startup's graph-based 'Persistent Intelligence' layer has convinced RRE Ventures and industrial giants to back a $35 million Series A.