PuppyGraph Raises $10M Series A After Sevenfold ARR Growth

PuppyGraph has raised a $10 million Series A led by Particle Future, with participation from defy.vc, after annual recurring revenue grew sevenfold over the past year. The company announced the round on November 10, 2026. The financing brings the graph query engine developer's total funding to $15 million.

The round follows the general availability of PuppyGraph 1.0. Customers include AMD, Palo Alto Networks, Coinbase, Trend Micro, Datadog and Sola Security. The company plans to spend the new capital on engineering expansion and enterprise go-to-market work.

The funding signals where enterprise data infrastructure is heading. AI agents increasingly investigate business questions across departments rather than simply generate SQL for a known dataset, and that shift strains traditional table-based architectures. PuppyGraph positions itself as the layer that lets those agents follow relationships without rebuilding the data stack around a dedicated graph database.

What Problem Does PuppyGraph Solve?

The core problem is relational sprawl. An agent investigating a business question may need to follow relationships among customers, products, components and suppliers. As tables multiply, schema descriptions consume more of the agent's context window, and the number of possible join paths grows.

PuppyGraph's answer is to query data in place. The engine applies graph computation directly to supported databases, warehouses and data lakes. Records stay in systems such as Apache Iceberg, Databricks, Snowflake, PostgreSQL and Microsoft Fabric. That approach avoids three costs at once: a dedicated graph database, a separate graph copy of the data, and the associated loading pipelines.

The design does not force an either-or choice between query styles. Applications can still use SQL for filters and aggregations while choosing graph queries for relationship-heavy questions.

How Does the Ontology Control Agent Behavior?

PuppyGraph anchors its approach in an ontology — a model of approved business entities and relationships that gives agents valid investigation paths. The company's AI-assisted schema builder proposes that model from existing tables, but data teams review and approve its business meaning before it governs agent behavior.

The engine actively enforces the ontology. It rejects references to entities or relationships absent from the model and returns structured feedback for query correction. PuppyGraph frames this as an answer to a specific failure mode in agent-driven analysis: silent errors caused by an agent inventing a connection and treating the resulting output as evidence.

The company is candid about the limits of that safeguard. Validation does not guarantee a correct conclusion, PuppyGraph cautions, because an agent can still pursue a poor hypothesis or conduct an incomplete analysis. The ontology constrains what paths exist; it does not judge whether the reasoning along those paths is sound.

Compatible AI tools connect through MCP, and users can also work through PuppyGraph's built-in agent.

How Fast Is It, and Why Does Speed Matter?

PuppyGraph reported that its engine supports subsecond graph queries for supported workloads and enables agents to query across 3,000 enterprise tables as one graph.

In a separate company benchmark, a four-machine cluster completed a 10-hop query across billions of edges in 2.26 seconds. The company notes that performance depends on workload.

The benchmark figure matters for a reason beyond raw throughput. Agents test multiple hypotheses in sequence, and a slow query can stall the next analytical step when it depends on the previous result. Deep relationship searches therefore need to remain interactive, or the agent's investigation loop breaks. A 10-hop traversal in 2.26 seconds on four machines is the kind of number that determines whether multi-department questioning stays practical or collapses into batch processing.

What Governance Controls Ship in Version 1.0?

Enterprise adoption of agent-driven querying hinges on permissions, and PuppyGraph 1.0 addresses that directly:

• Row-level permissions apply throughout graph traversals, including intermediate steps, so an agent cannot see restricted rows mid-query even if the final output is permitted.
• Role-based controls and user impersonation tie access to user identity.
• Single sign-on is supported.
• Administrative audit logs record management activity.
• Credential vending obtains short-lived, identity-aware credentials for platforms such as Snowflake and Databricks without requiring users to expose or manage long-lived credentials.

The intermediate-step enforcement is the notable detail. A graph traversal can pass through many records between its start and end points, and permissions checked only at the final result would leak information along the way. PuppyGraph applies the controls at every step.

What Do Customers Say?

Three named customers described production use cases.

Sola Security CTO Leon Goldberg said: "We run AI agents across thousands of tables and billions of edges a day."

Palo Alto Networks engineering director Ran Xia said the company runs multi-hop traversals on its existing data platform without moving or duplicating rows, eliminating the associated ETL pipeline maintenance. That use case goes to the core of PuppyGraph's pitch: the graph capability arrives without a new data pipeline to build and maintain.

Coinbase senior staff software engineer Xinyu Liu said Coinbase connects directly to tables maintained in Snowflake and Databricks and uses MCP for natural-language relationship queries.

AMD's deployment illustrates how adoption spreads. The company began with IT workloads and is expanding toward information-security and digital-marketing data, PuppyGraph reported. That progression matches PuppyGraph's broader model: as a relationship map connects additional departments, agents can investigate questions that previously stopped at organizational boundaries.

What Comes Next for the Company?

PuppyGraph plans to use the new capital for three stated priorities: reduce graph-model setup work, make relationship queries practical at enterprise scale, and help customers move agents into production.

The central challenge the company identifies is balance. Broadening agent investigations across more departments and more tables must not weaken the two control mechanisms that make those investigations trustworthy in the first place — the human-approved ontology that determines which relationships are valid, and the permission controls that determine which data each user can see. Scale the first without the second and agents gain reach without accountability; weaken the controls to speed deployment and the whole proposition of governed, in-place graph querying erodes.

The sevenfold ARR growth and the customer roster suggest the market has already validated the in-place approach. The $10 million round now tests whether PuppyGraph can turn that traction into production deployments at scale, where setup friction, permission fidelity and interactive query speed decide whether agents move from pilot projects into daily enterprise workflows.