PuppyGraph gets $10M for AI agents to query connected data

PuppyGraph, in Santa Clara, California, makes a graph query engine. This software shows connected records in many tables as one network, and it does not copy the data. On November 10 it got $10 million in Series A funding. Particle Future and defy.vc are the main investors. Total funding is at this time $15 million.

How it works

PuppyGraph reads tables in platforms such as Snowflake and Databricks. An AI tool makes a graph layout that teams examine before use. Agents then use English, or MCP, a standard link between agents and tools. Permissions for each row apply to each query.

The numbers

The company tells that a query of 10 links was complete across billions of edges in 2.26 seconds on four machines. It tells that its annual recurring revenue increased 7x in one year, but gave no revenue figures. The source shows no external test.

What is not known

Neo4j and TigerGraph keep data in graph databases that they make, a step that PuppyGraph tells companies can skip. The source tells that the question is if a query of data where it is can become the default. The customer quotes are from the company.

Insight: The fight is over where agent data lives

MariaDB puts agent memory and cache in one system, and its CEO tells that agents must have one data system below them. Google tells that its Knowledge Catalog does not move data. PuppyGraph also keeps data where it is. This reads as a race to hold the layer between agents and company data. The layer that agents use first sets the rules.