PuppyGraph has raised $10 million after releasing version 1.0 of its software, which lets AI agents query company data across thousands of tables without moving it into a separate graph database. The practical pitch is a way to connect an agent to scattered business data while keeping permissions and data relationships in view.
Watch Desk analysis
What happened
SiliconANGLE reports that PuppyGraph’s query engine can work across 3,000 enterprise tables, with the company saying graph results return in under a second. Teams can start with an AI-assisted schema draft based on existing tables, approve it, then ask questions in plain English using PuppyGraph’s agent or other tools through the Model Context Protocol.
The software rejects queries that refer to relationships not defined in the approved schema, rather than letting an agent guess how data fits together. When connected to Snowflake or Databricks, it can use short-lived credentials tied to a user’s identity; existing row-level permissions remain in force and, the company says, are checked at each step of a traversal. Developers can download a free version using Docker, with no sign-up, according to the report.
Particle Future and defy.vc led the Series A. PuppyGraph says its annual recurring revenue grew sevenfold over the past year, but disclosed no dollar figure. The report names AMD, Palo Alto Networks, Coinbase, Datadog and Trend Micro among its customers. Read SiliconANGLE’s report.
Why it matters
AI agents are only as useful as the information they can reach, and enterprise data is often spread across systems that were never designed to answer one joined-up question. PuppyGraph’s approach aims to let agents follow connections across existing warehouses and lakehouses without first building a separate graph database or handing over long-lived credentials.
The safety details are part of the product, not decorative trim: defined relationships constrain what the agent can query, while user-linked credentials and row-level permissions are intended to preserve existing access controls. Those are company claims reported by SiliconANGLE, not independent performance or security testing.
Our read
The interesting move is pairing natural-language questions with a structured map of what the data means and who is allowed to see it. That could make agents more useful in real company workflows, where an answer assembled across many systems is worth more than a clever demo.
The distinction to keep an eye on is between a promising architecture and dependable operation at scale. The company says it can handle thousands of tables quickly; customers will care whether it does so accurately, securely and without turning a data query into an unexpected tour of the entire business.
What to watch
- Whether PuppyGraph publishes independent performance or security evaluations.
- How organisations use the software beyond pilots, and what kinds of queries they trust it with.
- Whether the reported sevenfold revenue growth continues, and whether the company discloses a baseline.
Discussion spark: Would you let an AI agent query data across thousands of company tables if it could only follow approved relationships and the user’s existing permissions, or is that still too much access?
Sources and evidence
- PuppyGraph raises $10M to let AI agents query enterprise data as a graph – SiliconANGLE (10 November 2026, 14:00 UTC)
Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.