Edelbyte Exits Stealth With $70M to Unify Data Infrastructure for AI Agents

Edelbyte emerged from stealth on November 11, 2026 with $70 million in funding led by Sequoia and Quiet Capital, according to the company's official news listing. The startup is betting that AI agents will force companies to rebuild their data infrastructure around a single platform rather than a patchwork of specialized databases.

The round, announced the same day the company introduced itself publicly, also drew participation from Greenoaks, Swish Ventures, QP Ventures, Cerca Partners and Startpoint Capital. Edelbyte did not disclose a valuation.

What is Edelbyte selling?

The core pitch is consolidation. One platform handles search, analytical queries, vector, graph and geospatial work, removing the need for customers to assemble separate specialized databases for each task. The product overview places those functions alongside purpose-built tools for AI agents and governed access to detailed company data.

That last phrase carries the weight of the argument. Sequoia partner Bogomil Balkansky described the demand Edelbyte is targeting: "AI agents will ask far more questions of a company's data than people ever did."

The claim frames the market shift. Human analysts run bounded queries at human speeds. Autonomous agents, by contrast, can interrogate corporate data continuously, from every angle, without warning. Systems built for people — a search index here, a vector database there, a warehouse for analytics — were never designed for that access pattern. Edelbyte wants to replace the assembly with one engine.

Who built it?

The four co-founders are CEO Avner Aharoni, CTO Evgeney Ryzhyk, VP Engineering Alexander Sloutsky and VP Platform Engineering Yoni Leibowitz. All four are former Microsoft data-infrastructure experts. The company's about page describes years spent creating and operating systems for extremely large data workloads.

That pedigree matters in this segment. Data infrastructure is an engineering discipline where scale failures are expensive and public, and buyers weigh operational scars heavily. The founders' background in large-workload systems at Microsoft is the credential Edelbyte will lean on as it asks enterprises to consolidate mission-critical work on an unproven platform.

How does the architecture handle agent workloads?

The design follows directly from the Balkansky thesis. Comprehensive indexing is intended to handle ad hoc queries without requiring customers to anticipate every access pattern in advance. This is a pointed departure from traditional analytics, where engineers pre-model data for known questions. Agents do not file their questions in advance.

A built-in knowledge base gives agents context for their searches, so a query arrives with understanding of the data it is hitting rather than as a bare string. The company says this approach keeps detailed history accessible even when query volumes are high — a direct answer to the risk that agent-driven traffic overwhelms systems that trade history retention for query speed.

Edelbyte markets millisecond queries on petabytes of data and a single platform spanning hundreds of petabytes. Its example workloads sketch the intended customer base:

• Spotting suspicious logins and transactions as they happen
• Serving live figures in customer-facing applications
• Analyzing streaming device telemetry
• Querying complete logs and traces for observability

That list spans fraud detection, operational analytics and developer tooling — three markets usually served by three different categories of product. Edelbyte is asking for all of them on one engine.

Where can it run, and what does it connect to?

Deployment flexibility is a headline feature. Customers can run Edelbyte in public or sovereign clouds, self-hosted environments or on-premises, and query data where it resides. The sovereign-cloud and on-premises options signal an enterprise and regulated-industry focus, where data residency rules and security postures often block adoption of cloud-only platforms.

The listed data connections include:

• Apache Iceberg, Parquet and S3 for lakehouse and object storage
• Kafka for streaming ingestion
• PostgreSQL and MySQL for operational databases

For users and agents, integrations include Claude, Codex, Grafana, Tableau and Jupyter. The mix is telling: Claude and Codex cover the agentic side, Grafana and Jupyter cover engineering and data-science workflows, and Tableau covers conventional business intelligence. Edelbyte is positioning itself as the substrate beneath all of them rather than a competitor to any.

What does it cost?

The commercial model is pay as you go, without a commitment. On its pricing page, Edelbyte details the metering:

• Compute for queries and ingestion is billed per vCore-hour
• Hot cache is billed per GB-month, keeping data close to compute for millisecond queries
• Long-term storage is billed per GB-month, retaining full history at object-storage cost and remaining queryable on the same engine
• Networking and I/O operations are charged separately at cloud cost

The pricing structure mirrors the architecture. Hot cache buys speed; long-term storage buys completeness at object-storage economics. Because both remain queryable through one engine, customers do not have to choose between fast recent data and affordable full history — the traditional trade-off that forces teams to expire data or move it to slower, harder-to-query archives.

Usage-based pricing also lowers the barrier for a platform asking customers to consolidate. A company can pilot Edelbyte against one workload — observability logs, say — before migrating fraud detection or customer-facing analytics.

What about security and compliance?

The security features listed on the pricing page include encryption in transit and at rest, granular role-based access control, and single sign-on through SAML and OIDC. The page also displays compliance badges for SOC 2 Type II and ISO/IEC 27001.

Those credentials are table stakes for the enterprise buyers Edelbyte needs, but the granular access control carries extra weight in an agent world. If AI agents query company data autonomously, governing what each agent — and each human behind each agent — can see becomes a first-order security problem, not an afterthought. Edelbyte's pairing of agent tooling with governed access suggests the founders designed for that tension from the start.

Why does this round matter?

The $70 million raise lands in a market where the database category has spent a decade fragmenting. Search engines, vector databases, graph databases, time-series stores and cloud warehouses each carved out a specialty, and enterprises stitched them together with pipelines that break. The agent era pressures that model from both ends: agents ask questions no one modeled, and they ask them at volumes no human team generates.

Sequoia's involvement signals venture conviction that agent-driven data access is a platform opportunity rather than a feature bolted onto existing warehouses. Balkansky's quote is the thesis in one line: the demand curve for data queries is about to bend, and current infrastructure was not built for where it bends to.

Edelbyte enters with competition on every flank — established cloud warehouses adding AI features, vector-database vendors broadening their scope, and hyperscalers bundling search and analytics ever tighter. Its differentiation is the bet that consolidation, not specialization, wins the agent era.

What is the central test?

Edelbyte is asking customers to consolidate several kinds of data work on a system designed for both established analytics and agent-driven access. The central test for the new company is whether its usage-based service can deliver the advertised speed and cost efficiency at very large scale while supporting that breadth across cloud and customer-controlled environments.

Millisecond queries on hundreds of petabytes, billed by the vCore-hour, with no commitment — that is the promise enterprise buyers will now hold the company to. The four ex-Microsoft founders have the funding and the operating experience to attempt it; the market will decide whether one engine can truly serve every question an agent asks.