Santa Clara startup PuppyGraph has raised $10 million in a Series A round led by Particle Future with participation from defy.vc. The round brings the company’s total funding to $15 million and follows the general availability of PuppyGraph 1.0.
The company builds a graph query engine for enterprise AI agents. It lets agents explore data spread across thousands of tables as one connected graph, while the data stays in the warehouses and lakehouses where it already lives.
Why AI agents need a graph, not just SQL
Most enterprise AI tools today turn a question into SQL. That works when someone already knows which tables hold the answer and how they connect. It breaks down when an agent must discover those connections itself across thousands of tables.
PuppyGraph argues the fix is a graph model of the business. Entities such as customers, deals, products and suppliers are mapped once, with their relationships defined and approved by the data team. Agents then query that model directly.
The engine also guards against a quiet failure mode. If an agent references a relationship the model does not define, PuppyGraph rejects it and returns feedback the agent can use to correct the query. That stops an invented connection from becoming evidence.
The numbers behind the round
PuppyGraph says its annual recurring revenue grew sevenfold over the past year, though it has not disclosed the revenue figure. Its customers include AMD, Palo Alto Networks, Datadog, Netskope, Coinbase, Trend Micro, Baker Hughes and Sola Security.
The company also published a performance benchmark. Its engine completed a 10-hop query across billions of edges in 2.26 seconds on a four-machine cluster. Typical graph queries run in under a second, and the engine is built for petabyte-scale data.
AMD started with IT workloads and has since extended the platform into information security and digital marketing data. Other customers have moved from observability into fraud detection and from early agent workloads into new business areas.
A team built for query engines
PuppyGraph was co-founded by Weimo Liu, a computer science PhD from George Washington University. Liu previously worked on Google’s F1 query engine team and as a research scientist at TigerGraph.
His co-founders are Danfeng Xu, a nine-year veteran of LinkedIn’s infrastructure team, and Lei Huang, a three-time Google Code Jam world finalist. Zhenni Wu, a go-to-market veteran from the graph database sector, also joined the founding team.
The company launched in March 2024 and raised a $5 million seed round led by defy.vc in November 2024. defy.vc returned for the Series A, and the firm says PuppyGraph was built from day one for performance at scale.
“The insights of enterprise data will soon be generated by agents. Users will expect to ask questions that cross departments and systems, beyond the workflows someone built into a dashboard. That requires agents to navigate thousands of tables, follow the right relationships, and retrieve evidence in subseconds.”
Weimo Liu, cofounder and CEO, PuppyGraph
What the money will fund
The Series A will fund engineering and enterprise go-to-market expansion. The company’s priorities are reducing the work needed to build the graph model, making relationship queries practical at enterprise scale, and helping more teams put AI agents into production.
PuppyGraph 1.0 adds an AI-assisted schema builder that proposes entities and relationships from existing tables for teams to review and approve. Agents can reach it through MCP, and the platform enforces row-level permissions through every step of a graph traversal.
The wider bet
Traditional graph databases ask teams to load data into a separate store and keep it in sync. PuppyGraph takes the opposite route, bringing graph compute to the data where it already sits. Its supported sources include Apache Iceberg, Databricks, Snowflake, ClickHouse, DuckDB, PostgreSQL, Microsoft Fabric and Apache Hudi.
The bet is that agents, not dashboards, will do the next wave of enterprise analysis. If that holds, the company that lets agents investigate freely without moving data has a real edge. The sevenfold revenue growth suggests early customers agree, though the undisclosed revenue base leaves the scale of that growth unclear.
