Key Takeaways
- • PuppyGraph raised a $10 million Series A led by Particle Future and defy.vc, bringing total funding to $15 million, following the general availability of PuppyGraph 1.0.
- • Annual recurring revenue grew 7x over the past year, according to the company, with customers including AMD, Palo Alto Networks, Datadog, Netskope, Coinbase and Trend Micro.
- • PuppyGraph 1.0 queries enterprise data as one federated graph without loading it into a separate graph database, completing a 10-hop query across billions of edges in 2.26 seconds on a four-machine cluster.
- • The release targets the scaling problem of AI agents that must navigate roughly 3,000 tables, using an AI-assisted schema builder and MCP-compatible access.
- • Access controls include row-level permissions enforced during traversals, single sign-on, credential vending and administrative audit logs.
Summary
PuppyGraph, a graph query engine for enterprise AI agents based in Santa Clara, California, announced a $10 million Series A led by Particle Future and defy.vc. The round brings total funding to $15 million and will fund engineering and enterprise go-to-market expansion. The financing follows the general availability of PuppyGraph 1.0, and the company said annual recurring revenue grew 7x over the past year.
The company positions its product against a specific problem: enterprise AI agents that move from proof-of-concept work covering 30 to 50 tables to production questions spanning roughly 3,000 tables and multiple systems. According to the announcement, growing schema descriptions consume agent context windows, join paths multiply, and deep relationship queries can become too slow or time out. Moving data into a separate graph database adds loading delays, pipeline maintenance and scalability limits, which PuppyGraph argues its architecture avoids.
PuppyGraph 1.0 includes an AI-assisted graph schema builder that generates a schema from existing tables for teams to review and approve. Users can query data in plain English through a built-in agent or through compatible AI tools connected via MCP. The engine rejects undefined relationship references and returns feedback agents can use to correct their queries. The product supports Apache Iceberg, Databricks, Snowflake, ClickHouse, DuckDB, PostgreSQL, Microsoft Fabric and Apache Hudi, among other sources, and runs graph compute alongside SQL compute over the same data.
On security, the company says PuppyGraph 1.0 enforces configured row-level permissions throughout graph traversals, including intermediate steps, and supports single sign-on, role-based controls, user impersonation, short-lived credential vending for platforms such as Snowflake and Databricks, and administrative audit logs.
Customers quoted in the release include Ran Xia, Engineering Director at Palo Alto Networks, who said the product allows multi-hop traversals without moving or duplicating data, and Leon Goldberg, CTO of Sola Security, who said the company runs AI agents across thousands of tables and billions of edges daily. Weimo Liu, cofounder and CEO of PuppyGraph, said enterprise insights will increasingly be generated by agents. Amy Yin, partner at defy.vc, said the firm's investment rests on performance at scale. The company's figures, including the 7x ARR growth and 2.26-second benchmark, are self-reported and were not independently verified in the release.
The company positions its product against a specific problem: enterprise AI agents that move from proof-of-concept work covering 30 to 50 tables to production questions spanning roughly 3,000 tables and multiple systems. According to the announcement, growing schema descriptions consume agent context windows, join paths multiply, and deep relationship queries can become too slow or time out. Moving data into a separate graph database adds loading delays, pipeline maintenance and scalability limits, which PuppyGraph argues its architecture avoids.
PuppyGraph 1.0 includes an AI-assisted graph schema builder that generates a schema from existing tables for teams to review and approve. Users can query data in plain English through a built-in agent or through compatible AI tools connected via MCP. The engine rejects undefined relationship references and returns feedback agents can use to correct their queries. The product supports Apache Iceberg, Databricks, Snowflake, ClickHouse, DuckDB, PostgreSQL, Microsoft Fabric and Apache Hudi, among other sources, and runs graph compute alongside SQL compute over the same data.
On security, the company says PuppyGraph 1.0 enforces configured row-level permissions throughout graph traversals, including intermediate steps, and supports single sign-on, role-based controls, user impersonation, short-lived credential vending for platforms such as Snowflake and Databricks, and administrative audit logs.
Customers quoted in the release include Ran Xia, Engineering Director at Palo Alto Networks, who said the product allows multi-hop traversals without moving or duplicating data, and Leon Goldberg, CTO of Sola Security, who said the company runs AI agents across thousands of tables and billions of edges daily. Weimo Liu, cofounder and CEO of PuppyGraph, said enterprise insights will increasingly be generated by agents. Amy Yin, partner at defy.vc, said the firm's investment rests on performance at scale. The company's figures, including the 7x ARR growth and 2.26-second benchmark, are self-reported and were not independently verified in the release.
Go deeper than the headline
Research every firm in this story, track the ones that matter and get the daily briefing — free for 7 days.