PuppyGraph
A graph query engine for enterprise AI agents that queries existing relational data without ETL.
Website: https://www.puppygraph.com/
Cover Block
Open sources
| Field | Value |
|---|---|
| Name | PuppyGraph |
| Tagline | A graph query engine for enterprise AI agents that queries existing relational data without ETL [PuppyGraph] |
| Headquarters | Santa Clara, US [Crunchbase] |
| Founded | 2024 [Crunchbase] |
| Stage | Series A [Axios Pro, November 2026] |
| Business model | SaaS |
| Industry | Deeptech |
| Technology | AI / Machine Learning |
| Geography | North America |
| Growth profile | Venture Scale |
| Founding team | Co-Founders (3+), Weimo Liu, Danfeng Xu, Lei Huang [Business Wire, November 2024] |
| Funding label | $10M+ |
| Total disclosed funding | ~$15,000,000 [Axios Pro, November 2026] |
Links
Open sources
- Website: https://www.puppygraph.com/
- LinkedIn: https://www.linkedin.com/company/puppygraph
- Careers: https://puppysphere.applytojob.com/
What an Investor Needs First
PUBLIC PuppyGraph is building a graph query engine that lets enterprises query relationships in existing relational and lakehouse data without moving that data first, a proposition that is getting investor attention because the company is tying a long-standing data infrastructure problem to the newer demand for AI agents with connected context [PuppyGraph] [Axios Pro, November 2026] [SiliconANGLE, November 2026]. Founded in 2024 and launched in March of that year according to its seed announcement, the Santa Clara company argues that customers can expose existing tables as a graph layer rather than stand up and maintain a separate graph database, with PuppyGraph 1.0 reaching general availability alongside its Series A announcement in November 2026 [Business Wire, November 2024] [PuppyGraph] [Blocks & Files, November 2026].
The product differentiation, at least in public materials, rests on deployment simplicity and performance claims: PuppyGraph says it can be deployed to first query in 10 minutes and operate as a zero-ETL graph analytics layer on top of warehouses and lakehouses already in place [PuppyGraph]. Third-party coverage and company-linked benchmark references also point to multi-hop query performance across very large graph datasets, although investors should treat the strongest speed and scale figures as benchmark evidence rather than audited production telemetry [BigDATAwire, November 2024] [Venture Capital Access Online, November 2024] [iTechGuides].
The founding team looks technically credible for this category. CEO Weimo Liu previously worked on Google’s F1 database team and at TigerGraph, co-founder Danfeng Xu spent nine years on LinkedIn’s infrastructure team, and co-founder Lei Huang is identified in public materials as a multiple-time Google Code Jam finalist, a mix that suggests depth in databases, distributed systems, and query performance rather than a thin application-layer team [Business Wire, November 2024] [Data Learning Science] [LinkedIn, January 2026].
On financing, PuppyGraph has disclosed $15 million across a $5 million seed led by defy.vc in November 2024 and a $10 million Series A led by Particle Future in November 2026, with a SaaS business model and stated plans to use the latest round for engineering and enterprise go-to-market expansion [Business Wire, November 2024] [Axios Pro, November 2026]. Public traction signals are promising but still need careful handling: the company has cited production customers and said ARR grew sevenfold year over year, yet those commercial claims are still largely company-reported in the available record [SiliconANGLE, November 2026] [Blocks & Files, November 2026]. Over the next 12 to 18 months, the key watchpoints are whether PuppyGraph can convert technical interest into repeatable enterprise adoption, broaden proof points beyond company-stated customer references, and show that its zero-ETL graph layer becomes durable infrastructure for AI-agent workflows rather than a point solution attached to current agent experimentation [PuppyGraph] [Axios Pro, November 2026] [SiliconANGLE, November 2026].
Claim stands unchecked -- This section relies materially on company statements and company-reported metrics, with partial corroboration from Axios Pro, SiliconANGLE, Blocks & Files, and Business Wire.
Taxonomy Snapshot
| Axis | Value |
|---|---|
| Stage | Series A |
| Business Model | SaaS |
| Industry / Vertical | Deeptech |
| Technology Type | AI / Machine Learning |
| Geography | North America |
| Growth Profile | Venture Scale |
| Founding Team | Co-Founders (3+) |
| Funding | $10M+ total disclosed, approximately $15,000,000 |
Inside the Company
PUBLIC PuppyGraph presents a fairly specific infrastructure bet: treat the enterprise data stack as fixed, and add a graph query layer on top rather than asking customers to stand up a separate graph database. Public company materials describe it as a Santa Clara-based startup founded in 2024, with a product that queries existing relational data stores as a graph without ETL or data movement [Crunchbase] [PuppyGraph]. The same materials position the company around enterprise AI-agent workflows as well as graph analytics use cases, which is useful context because it frames PuppyGraph less as a system-of-record database vendor and more as an access and query layer on top of warehouses and lakehouses [PuppyGraph].
The chronology that is publicly supportable is short but coherent. Crunchbase lists PuppyGraph as founded in 2024 and headquartered in Santa Clara, California [Crunchbase]. On its website, the company describes PuppyGraph as "The Graph Engine for Your Existing Data" and says customers can query tables as a graph in under 10 minutes, with no graph database required [PuppyGraph]. By November 2024, the company had announced a $5 million seed round led by defy.vc, and by November 2026 it had announced a $10 million Series A led by Particle Future, bringing disclosed funding to $15 million [Crunchbase]. That sequence suggests a company moving from initial product formation into a more formal enterprise commercialization phase, although public materials here do not establish the legal entity name or a fuller operating history [Crunchbase] [PuppyGraph].
Verified against public records -- Confirmed by Crunchbase and company website.
Under the Hood
MIXED
PuppyGraph is selling a specific architectural claim, not a general AI workflow tool. The product is described as a graph query engine that sits on top of existing relational stores, warehouses, and lakehouses, then exposes that data as a graph without requiring ETL or a separate graph database [PuppyGraph] [Business Wire, November 2024]. On its homepage, the company says customers can "query your tables as a graph," enable agents across "3,000 tables," and reach a unified graph model in under 10 minutes, while avoiding data movement [PuppyGraph]. Press coverage around the November 2026 Series A presents the same basic positioning, tying the product to analytics, security, fraud, and enterprise AI agent use cases [SiliconANGLE, November 2026] [Blocks & Files, November 2026].
The technical differentiation, as publicly described, rests on query execution over data that already lives elsewhere. PuppyGraph says it can run direct graph analysis on platforms such as Databricks and Delta Lake without copying data, and VMblog reported that the company became a design partner for Databricks Unity Catalog in June 2025 [PuppyGraph] [VMblog, June 2025]. The product surface also appears broader than a core query layer alone: the company offers a free Developer tier, an Enterprise tier with a 30-day trial, and deployment options via AWS AMI and Docker [PuppyGraph Pricing]. InfoWorld reported that PuppyGraph 1.0 added an AI-assisted graph schema builder and MCP integrations, which suggests the company is trying to reduce setup friction for agent and analytics workloads rather than competing only on raw graph performance [InfoWorld].
Public performance evidence is directionally encouraging but still thin by institutional standards. The company homepage claims subsecond querying at petabyte scale and a footprint across thousands of tables [PuppyGraph]. Separately, third-party writeups cited a six-hop query across 600 million edges in under one second and a 10-hop neighbor query across half a billion edges in 2.26 seconds, with one version specifying a four-node cluster [iTechGuides] [BigDATAwire, November 2024] [Venture Capital Access Online, November 2024]. Those figures indicate real engineering ambition, but most of the strongest operating claims remain company-originated or reproduced from company materials, so the main diligence question is less whether the engine works at all and more how consistently those benchmarks hold across messy enterprise schemas and production security constraints.
Reasoned from indirect evidence -- This section combines two or more independent public reports with several material product and performance claims that remain company-originated or only lightly corroborated.
Market Research
PUBLIC
The market matters now because enterprise AI adoption is colliding with a stubborn data-layout problem: most large companies still store operational context in relational tables and lakehouse formats, while many high-value agent, security, and fraud workflows depend on traversing relationships rather than reading rows in isolation [SiliconANGLE, November 2026] [PuppyGraph] [Blocks & Files, November 2026].
The public record does not provide a direct third-party TAM, SAM, or SOM estimate for PuppyGraph’s exact category, which is best described as zero-ETL graph query infrastructure for enterprise data and AI agents. What is visible is the shape of the category the company is trying to enter. PuppyGraph positions itself as a graph query engine layered on top of existing warehouses and lakehouses, with use cases spanning knowledge graph analytics, cybersecurity, supply chain, fraud, and healthcare [PuppyGraph]. Named coverage from SiliconANGLE and Blocks & Files repeats the same basic market framing, namely that enterprises want graph-style querying without first building and maintaining a separate graph database or moving data into one [SiliconANGLE, November 2026] [Blocks & Files, November 2026]. That places PuppyGraph at the intersection of several adjacent public markets rather than a single clean category.
The nearest public analogs are the graph database and graph analytics market on one side, and the broader data-lakehouse and AI infrastructure market on the other. A narrower public sizing table is more honest here than a synthetic TAM. It shows that the opportunity is being created by adjacent markets with established spend pools, not by a category that is already fully standardized in public market research.
| Analogous market | Public sizing claim | Relevance to PuppyGraph |
|---|---|---|
| Graph database market | $3.4 billion by 2030 (analogous market) [Grand View Research, January 2023] | Captures enterprise demand for relationship-centric querying and analytics |
| Knowledge graph market | $4.8 billion by 2030 (analogous market) [MarketsandMarkets, October 2023] | Relevant to semantic enterprise context and connected-data applications |
| Data lake market | $34.1 billion by 2030 (analogous market) [Grand View Research, July 2023] | Relevant because PuppyGraph sits on top of existing lakehouse and warehouse data rather than replacing it |
| Generative AI in enterprise market | $67.2 billion by 2030 (analogous market) [MarketsandMarkets, May 2024] | Relevant because PuppyGraph’s current messaging is increasingly tied to AI agents and connected context |
The sizing table points to a useful distinction. PuppyGraph does not need to win a standalone graph database budget line to matter; it can also attach to lakehouse modernization and agent-infrastructure budgets if buyers accept graph querying as an access layer rather than a destination store.
Demand drivers in the cited research are practical rather than theoretical. PuppyGraph’s homepage and product materials argue that enterprises want to query existing relational data as a graph, in less than 10 minutes to first query according to the company, and without copying data or remodeling schemas [PuppyGraph]. External coverage links that proposition to agent workflows specifically, with Axios Pro, SiliconANGLE, and Blocks & Files all framing the Series A around helping AI agents query enterprise data as a graph [Axios Pro, November 2026] [SiliconANGLE, November 2026] [Blocks & Files, November 2026]. If that framing holds in production, the tailwind is straightforward: agent systems need more connected context than keyword retrieval alone can provide, and enterprises are reluctant to create another governed data store just to get it.
The adjacent and substitute markets are equally important because they define both upside and resistance. On the substitute side, enterprises can approximate parts of PuppyGraph’s value with native SQL on warehouses, search and vector retrieval layers, data catalog products, or conventional graph databases that require ETL into a separate engine [PuppyGraph] [SiliconANGLE, November 2026]. On the adjacent side, PuppyGraph’s design-partner status for Databricks Unity Catalog suggests a route into existing lakehouse governance and metadata workflows, which is more commercially promising than trying to force a greenfield graph migration in a mature enterprise account [VMblog, June 2025]. The presence of AWS AMI and Docker distribution also suggests the company is trying to meet infrastructure teams where they already buy and deploy software, not create a wholly new procurement motion [PuppyGraph Pricing].
Macro and regulatory forces cut in both directions. Data-governance pressure generally favors architectures that query data in place, since moving sensitive data across systems can increase security, lineage, and access-control complexity; that is consistent with PuppyGraph’s repeated no-data-movement message [PuppyGraph] [PuppyGraph For Cybersecurity]. At the same time, security and compliance teams usually require proof that a new query layer respects existing permissions and audit controls, especially if it is being used by AI agents. Public materials point to security and deployment as part of the product story, but they do not yet provide enough independent evidence to conclude that governance is a finished market advantage rather than a diligence item [PuppyGraph] [Blocks & Files, November 2026].
The market, then, looks less like a single category ready for share capture and more like a convergence zone. If enterprises continue to fund AI-agent infrastructure, while resisting fresh ETL-heavy data projects, the conditions are favorable for a graph access layer that can sit on top of existing systems. The open question is whether that demand settles into a durable software category or remains a feature expectation absorbed by broader data platforms.
Partially corroborated -- Section relies on named publisher coverage and company materials for category framing, with third-party analogous market reports used for sizing rather than a direct market definition.
Competition and Substitutes
MIXED PuppyGraph is positioning itself less as a new system of record and more as a query layer that lets enterprises treat existing warehouse and lakehouse data as a graph, which places it in competition with graph databases on one side and SQL-first data platforms on the other [PuppyGraph] [Blocks & Files, November 2026].
The competitive map is clearest when split by architecture. On one flank are purpose-built graph databases and graph analytics vendors, a category PuppyGraph implicitly challenges by arguing that enterprises should not have to build and maintain a separate graph database to run relationship queries on operational or analytical data [PuppyGraph] [Blocks & Files, November 2026]. On the other flank are incumbent cloud data platforms and lakehouse tooling, where the substitute is not another graph engine so much as doing relationship analysis with SQL, custom pipelines, or feature extensions inside existing platforms. PuppyGraph's Databricks Unity Catalog design-partner status matters here because it suggests the company is trying to attach itself to an already adopted control plane rather than force a rip-and-replace motion [VMblog, June 2025].
The company's edge today appears to rest on product architecture and founder-market fit, but both are still somewhat perishable. The architecture claim is straightforward: PuppyGraph says it can query existing relational and lakehouse data as a graph without ETL or data movement, with deployment to first query in 10 minutes and production support at 5 to 10 PB, all of which, if borne out in buyer diligence, reduce implementation friction relative to a dedicated graph stack [PuppyGraph]. The team also reads as technically native to the problem. CEO Weimo Liu previously worked on Google's F1 database team and at TigerGraph, Danfeng Xu spent nine years on LinkedIn's infrastructure team, and co-founder Lei Huang has public credentials in high-performance algorithmic work [Business Wire, November 2024] [Data Learning Science]. That is a credible recipe for early product advantage, but not yet a locked channel, proprietary dataset, or regulatory moat.
The exposure is equally clear. If Databricks, Google Cloud, or another large data-platform vendor closes the gap with native graph-query features around governed warehouse data, PuppyGraph could find itself competing against a control point customers already pay for and trust [VMblog, June 2025]. The same risk applies, differently, against established graph vendors: a specialist with deeper graph-native workflows, larger field organizations, or a longer record in enterprise production could still win accounts where the buyer wants a full graph database and not a zero-copy query layer. Public sources here do not name those rivals directly, so the comparison has to stay architectural rather than logo-specific.
The most plausible 18-month scenario is a split market. Databricks looks like the winner if graph analysis becomes a standard lakehouse feature set and customers prefer to consolidate on existing data-platform vendors, especially given PuppyGraph's own effort to integrate with that ecosystem [VMblog, June 2025]. PuppyGraph is the winner if buyers continue to want graph capabilities without data movement and without standing up another database tier, particularly in AI-agent, security, and fraud workflows where relationship traversal is valuable but deployment complexity is a buying obstacle [PuppyGraph] [SiliconANGLE, November 2026]. The likely loser if that second condition fails is any independent middleware layer that depends on technical superiority alone before it has durable distribution, and PuppyGraph still fits that description in public evidence.
Opportunity
PUBLIC
If PuppyGraph executes, the prize is not a niche graph-tools business but a control point in how enterprise AI agents retrieve connected context from the relational data estates companies already have, a layer that could sit in front of warehouses and lakehouses rather than asking buyers to rebuild them first [PuppyGraph] [SiliconANGLE, November 2026] [Blocks & Files, November 2026].
The headline opportunity is straightforward: become the default graph query layer for AI, security, and analytics workloads that need relationship reasoning on top of existing enterprise data. That outcome is reachable, not merely aspirational, because the company is selling into a real infrastructure constraint already visible in its public materials: large enterprises store relevant entities and relationships across thousands of tables, and PuppyGraph's pitch is that those tables can be queried as a graph without ETL or a separate graph database buildout [PuppyGraph] [Business Wire, November 2024]. The public evidence also shows early signs of fit with enterprise-grade environments rather than only developer experiments, including production customer claims spanning AMD, Palo Alto Networks, Datadog, Netskope, Coinbase, Trend Micro, and Baker Hughes, plus a design-partner relationship around Databricks Unity Catalog that points to alignment with the modern lakehouse stack [SiliconANGLE, November 2026] [VMblog, June 2025].
There are several credible paths from that wedge to much larger scale. The common thread is that PuppyGraph does not need to replace a warehouse, lakehouse, or operational database to matter; it needs to become the graph-access layer buyers standardize on when AI agents and analysts require multi-hop reasoning across existing systems [PuppyGraph] [Blocks & Files, November 2026].
| Scenario | What happens | Catalyst | Why it's plausible |
|---|---|---|---|
| Lakehouse graph layer | PuppyGraph becomes a standard add-on for Databricks and adjacent data platforms when enterprises want graph reasoning on existing lakehouse data | The Databricks Unity Catalog design-partner relationship expands into repeatable field adoption patterns [VMblog, June 2025] | The product is already positioned as zero-ETL graph analytics on existing data stores, and the design-partner signal suggests platform-level relevance rather than a standalone point tool [PuppyGraph] [VMblog, June 2025] |
| Agent infrastructure standard | Enterprise AI teams adopt PuppyGraph as the relationship-retrieval engine behind internal agents that need connected enterprise context | General availability of PuppyGraph 1.0, plus MCP integrations and AI-assisted schema building, lowers deployment friction for agent builders [Blocks & Files, November 2026] [InfoWorld] | The company explicitly targets AI-agent use cases, and its public materials frame the product around enabling agents across large table estates without remodeling the data first [PuppyGraph] [SiliconANGLE, November 2026] |
| Security and fraud expansion | Cybersecurity and risk teams become the first large-budget repeat buyer cohort, then expand usage into adjacent graph-heavy workloads | Category-specific messaging and proof points around multi-hop analysis on existing lakehouse and warehouse data improve time-to-value for security buyers [PuppyGraph For Cybersecurity] [Business Wire, November 2024] | Security, fraud, and attack-path analysis are already named use cases, and early customer references include security-oriented enterprises such as Palo Alto Networks, Netskope, Trend Micro, Sola Security, and Prevalent AI [PuppyGraph] [SiliconANGLE, November 2026] |
The compounding mechanism here is adoption without data relocation. If PuppyGraph can deliver useful graph queries on top of systems buyers already run, each deployment can start with one urgent workflow, then widen across more teams, more tables, and more use cases because the hard part, integrating with the existing data estate, is already done once [PuppyGraph] [Business Wire, November 2024]. The early signs of that flywheel are visible in product packaging and ecosystem posture: a free Developer tier can widen top-of-funnel experimentation, enterprise deployment is available through AWS AMI and Docker, and the company has discussed integrations with visualization and streaming tools that could make it easier for one successful use case to pull in adjacent workloads [PuppyGraph Pricing] [Insights from Analytics]. Public performance claims, including multi-hop queries across hundreds of millions to billions of edges in seconds, also matter here because they speak directly to whether a pilot can graduate into broader production use at enterprise scale [BigDATAwire, November 2024] [Venture Capital Access Online, November 2024].
The size of the win is best framed as infrastructure value capture rather than a simple seat-based SaaS outcome. There is no cited third-party TAM in the source set, and no direct public comparable is named in the research, so the cleaner way to bound upside is strategic importance: if PuppyGraph becomes the default graph-query layer for lakehouses and agent stacks, it could plausibly support a multibillion-dollar outcome (scenario, not a forecast) because it would sit on top of existing enterprise data spend rather than compete for a narrow new budget line [SiliconANGLE, November 2026] [Blocks & Files, November 2026]. That upside case is helped by two public facts investors tend to care about in infrastructure markets: the company reached production references quickly after its March 2024 launch, according to its seed announcement, and by November 2026 it had raised a Series A while claiming sevenfold ARR growth over the prior year, which at minimum suggests commercial momentum even if the revenue figure itself remains company-reported [Business Wire, November 2024] [Axios Pro, November 2026].
Claim stands unchecked -- This section relies partly on company materials and re-reported company claims for product capability, customer adoption, and ARR growth, with some corroboration from Axios Pro, SiliconANGLE, Blocks & Files, Business Wire, and VMblog.
Sources
Open sources
[PuppyGraph] PuppyGraph: The Graph Engine for Your Existing Data. No ETL. | https://www.puppygraph.com/
[Crunchbase] PuppyGraph - Crunchbase Company Profile & Funding | https://www.crunchbase.com/organization/puppygraph
[Axios Pro, November 2026] Exclusive: PuppyGraph raises $10M Series A | https://www.axios.com/pro/enterprise-software-deals/2026/11/10/puppygraph-query-defy-venture
[SiliconANGLE, November 2026] PuppyGraph raises $10M to let AI agents query enterprise data as a graph | https://siliconangle.com/2026/11/10/puppygraph-raises-10m-to-let-ai-agents-query-enterprise-data-as-a-graph/
[Business Wire, November 2024] PuppyGraph Raises $5 Million in Seed Funding Led by Defy.vc | https://www.puppygraph.com/blog/puppygraph-raises-5-million-in-seed-funding-led-by-defy-vc
[Blocks & Files, November 2026] PuppyGraph gets A-round funding to develop graph help for agents | https://www.blocksandfiles.com/ai-ml/2026/11/11/puppygraph-gets-a-round-funding-to-develop-graph-help-for-agents/5309071
[BigDATAwire, November 2024] PuppyGraph Announces Seed Funding and Benchmark Results for Zero-ETL Graph Query Engine | https://www.bigdatawire.com/2024/11/14/puppygraph-announces-seed-funding-and-benchmark-results-for-zero-etl-graph-query-engine/
[Venture Capital Access Online, November 2024] PuppyGraph Announces Seed Funding Led by defy.vc and Benchmark Results for Zero-ETL Graph Query Engine | https://www.vcaonline.com/news/2024/11/14/puppygraph-announces-seed-funding-led-by-defy-vc-and-benchmark-results-for-zero-etl-graph-query-engine/
[LinkedIn, January 2026] Jin Sha - Staff Software Engineer at LinkedIn | https://www.linkedin.com/in/jin-sha-26bb5689/
[Data Learning Science] Episode 297 Lei Huang Co-Founder @ PuppyGraph | https://www.datalearningscience.com/podcast/episode-297-lei-huang-cofounder-puppygraph
[VMblog, June 2025] Databricks Brings Graph Analytics to Unity Catalog with PuppyGraph | https://vmblog.com/archive/2025/06/11/databricks-brings-graph-analytics-to-unity-catalog-with-puppygraph.aspx
[Grand View Research, January 2023] Graph Database Market Size, Share & Trends Analysis Report | https://www.grandviewresearch.com/industry-analysis/graph-database-market-report
[MarketsandMarkets, October 2023] Knowledge Graph Market by Type, Technology, Organization Size, Application, Vertical and Region | https://www.marketsandmarkets.com/Market-Reports/knowledge-graph-market-48424377.html
[Grand View Research, July 2023] Data Lake Market Size, Share & Trends Analysis Report | https://www.grandviewresearch.com/industry-analysis/data-lake-market-report
[MarketsandMarkets, May 2024] Generative AI in Enterprise Market by Offering, Model Type, Deployment Mode, Business Function, End User and Region | https://www.marketsandmarkets.com/Market-Reports/generative-ai-in-enterprise-market-143759486.html
Articles about PuppyGraph
- PuppyGraph’s Zero-ETL Engine Queries 3,000 Enterprise Tables for AI Agents — The company’s graph query engine, now in GA, connects directly to data lakes to serve AI agents without moving petabytes of enterprise data.