PuppyGraph Raises $10M Series A for Enterprise AI Agents

An AI-assisted schema builder can suggest how an enterprise's records connect. At PuppyGraph, the data team reviews and approves that model before agents use it to investigate the business. That puts a familiar responsibility inside an unfamiliar workflow: someone has to decide which relationships deserve to become evidence.

The Santa Clara, California company announced a $10M Series A on November 10, 2026, bringing reported total funding to $15M. Particle Future and defy.vc co-led the financing, according to the distributed company release and the investor's own announcement. The proceeds will expand engineering and enterprise go-to-market work as AI applications reach further into existing company data.

That expansion gives data-platform owners a consequential place in the agent business. PuppyGraph sells a graph query engine that works over existing stores, allowing an investigation to follow connected records while the enterprise keeps its data infrastructure. The sales conversation therefore includes the people who understand those records, maintain access rules and will inherit the system after the demonstration ends.

What PuppyGraph's Series A funds

The funding follows the general availability of PuppyGraph 1.0 and an earlier $5M seed round led by defy.vc. The company published that seed announcement in November 2024. The new financing adds engineering and commercial capacity around a product whose adoption depends on both technical execution and the buyer's understanding of its own information.

CEO Weimo Liu brings experience from Google's F1 query-engine team and TigerGraph. CTO Danfeng Xu previously worked on LinkedIn infrastructure, while Chief Architect Lei Huang worked on payments systems at Google and Instacart. Co-founder Zhenni Wu leads go-to-market. Crunchbase and CB Insights also identify Kunjie Chen as a co-founder.

Those backgrounds help explain the engineering assignment without settling the commercial outcome. A graph engine has to make the connection between two records useful at the speed and scale a customer needs. Its seller also has to explain where that capability belongs in a platform the customer already operates. PuppyGraph's new capital funds more work on both sides of that conversation, with no disclosed valuation or absolute revenue figure to supply a shortcut to understanding the business.

Where the enterprise supplies the meaning

A database column tells software how information is arranged; its business meaning can demand a separate conversation. Two identifiers can look compatible while referring to different things. An agent allowed to keep investigating from a mistaken connection can carry that error into later questions, even when each individual query executes successfully.

PuppyGraph's account of its agent architecture gives the data team responsibility for approving entities and relationships proposed by the schema builder. The engine rejects references to relationships outside that model and returns feedback the agent can use to revise its query. This creates a defined place to review the connections available to an investigation, while leaving the quality of the business definitions consequential.

The division of work matters to an operator evaluating the product. A team could know where a table lives and still need a finance, security or operations colleague to explain what a record represents. Turning that knowledge into an approved relationship makes it available beyond the individual who happened to know the answer. It also gives the enterprise something concrete to revisit when the business changes.

That is a practical reason to take the modeling step seriously. An organization buying an agent's ability to explore has an interest in the map that guides the exploration. PuppyGraph's opportunity grows with the usefulness of that map to the next question the business asks.

A graph engine beside the existing data platform

PuppyGraph applies graph computation to supported databases, warehouses and lakehouses, including Databricks, Snowflake, Apache Iceberg and PostgreSQL. The company describes graph queries and SQL operating alongside each other over existing data. That architecture can let a team add relationship analysis without creating a separate graph copy and the loading work needed to keep it current.

There are several architectural jobs inside the graph market. Neo4j's documentation describes a native graph database with graph storage, transactions and an extensive tool ecosystem. Microsoft's GraphRAG project describes extracting a knowledge graph from text and using graph structures and summaries during retrieval. These approaches address different layers of the work, so a buyer needs more than the shared word “graph” to compare them sensibly.

For PuppyGraph, the commercial distinction is its relationship to the existing data estate. The warehouse or lake remains an operating asset with owners, budgets and controls. Extending that asset can change the scope of a deployment and the colleagues who need to approve it. Any claim about lower total cost or superior performance across all workloads would require evidence beyond the architecture itself.

The company also supports agent access through MCP and describes configured row-level permissions applying through graph traversals. The ability to follow a connection creates another access decision at the next step. Keeping those decisions attached to user identity helps the data team assess an investigation as part of its existing responsibility for the platform.

Expansion follows the questions customers bring

PuppyGraph reports 7x annual recurring revenue growth over the preceding year and names AMD, Coinbase, Datadog, Palo Alto Networks and other enterprises as customers. Those figures and customer relationships are company-reported. The growth rate comes without an absolute ARR figure, and published query-performance examples depend on their workloads and configurations.

Customer accounts on PuppyGraph's website describe querying data already maintained in enterprise platforms. For a platform team, that makes the introduction of graph analysis a question about the work surrounding an existing investment. Who will define the relationships, which questions should use them, and how will access behave when an investigation passes into another part of the business?

Those questions give the engineering and enterprise sales expansion a shared purpose. The engine must retrieve useful connections, while the deployment must fit the people and systems responsible for them. A new department can bring an additional use case and another set of definitions worth getting right.

The Series A puts more resources behind that effort. As PuppyGraph reaches further into an enterprise, the next useful relationship will have both a technical definition and a colleague who can explain why it belongs there.