OliverAI Raises Pre-Seed Funding for Agent-Native Data

OliverAI has announced undisclosed pre-seed funding from Menlo Ventures and Unusual Ventures to scale OliverDB, an analytical data platform designed for AI-agent workloads. The financing was announced on August 26, 2026, and the company did not disclose the amount, valuation, ownership terms, or a lead investor.

The round matters because AI agents change the rhythm of enterprise analytics. A human can pause between queries, review a dashboard, and decide what to ask next, while an agent can generate continuous analytical work at machine speed. OliverAI is betting that performance, compute economics, data permissions, and auditability now belong in the same infrastructure decision.

What OliverAI Announced

The official financing announcement identifies Praneet Sharma, Grace Johnson, and Toby O'Brien as OliverAI's co-founders. It says Menlo Ventures and Unusual Ventures provided pre-seed funding to scale OliverDB. Public materials do not assign CEO or CTO titles to the founders, and they do not disclose the round size, prior financing, total funding, valuation, or additional investors.

The intake headline described the event as Seed funding, but both the company announcement and Unusual Ventures' public confirmation use Pre-Seed. That primary-source label controls the published subcategory. It also fits the available company record: Oliver AI's LinkedIn page lists a 2026 founding year and a team of 2-10 people.

The only stated use of funds is to scale OliverDB. OliverAI says the platform is available for enterprise deployments as a managed service or inside a customer's VPC, and its current site also presents dedicated and serverless cloud options. A detailed hiring plan, geographic expansion plan, product budget, and go-to-market allocation have not been disclosed.

Why Agent Workloads Change the Database Problem

Traditional analytical systems were built around people and applications that create relatively bounded demand. A person submits a query or loads a dashboard, then time passes before the next request. An AI agent can investigate a problem, generate follow-up questions, call tools, revise a hypothesis, and repeat the process without the same pauses, turning a useful workflow into a much denser infrastructure load.

OliverAI's argument is that the database cannot solve that change through speed alone. If an agent receives fast access to sensitive data without a precise policy boundary, the enterprise gains latency while losing control. If the system adds governance but requires expensive compute for every analytical loop, the deployment can become difficult to justify before the agent reaches production value.

That is the combined buying problem behind the financing. Menlo Ventures has spent years investing across enterprise software, data infrastructure, and AI, while Unusual Ventures describes its own focus as early-stage enterprise AI software. Their participation places OliverAI inside an investment thesis that infrastructure for agents will be designed around machine users, rather than added as another feature to systems optimized for human analysts.

How OliverDB Is Designed

The OliverDB Engine is described as a columnar, cellular analytical database that can run as a small embedded binary or as a service. Durable data can remain in shared object storage, while stateless cells handle query execution without requiring a traditional sharding or rebalancing project. The engine supports SQL, a JSON query language, and full-text search.

OliverAI also says the platform can continuously ingest from systems including Postgres, Snowflake, ClickHouse, and Databricks. That matters commercially because an enterprise is more likely to test a new data layer when the evaluation does not begin with replacing every existing system. The product pitch is an agent-ready layer that can coexist with the current estate while giving agents a faster and more controlled path to analytical data.

Agent Flight Control handles the control boundary. OliverAI says each agent receives a policy that can restrict sources, tables, rows, columns, operators, writes, tool calls, and egress. The system rewrites SQL queries and MCP tool calls before execution, records allowed and denied actions, and keeps policy changes subject to customer approval rather than letting the reasoning layer move its own fence.

The OliverDB Swarm addresses a different part of the economics. Smaller models test competing hypotheses against enterprise data, share grounded findings, and return evidence to a conductor model that can weigh the result or conclude that the evidence is insufficient. The intended effect is broader investigation with less reliance on expensive frontier models for every step.

What the Benchmark Claims Actually Say

OliverAI's current benchmark notes report up to 162x CPU performance, up to 1000x GPU performance, and 50-300x lower compute use in evaluated analytical workloads. One internal comparison records an 8 ms query against 3.8 seconds on the same data. These numbers are company-reported, directional, workload-dependent, and explicitly not independently audited.

The qualification belongs beside the numbers because database performance changes with query shape, joins, cardinality, storage temperature, concurrency, hardware, caching, and deployment topology. OliverAI says technical evaluations use a customer's dataset, query suite, hardware, and measurement window, with methodology and raw results shared for that workload. That is the right place for the technical claim to become a commercial one: inside a reproducible customer evaluation, not in a universal speed slogan.

What Menlo and Unusual Are Underwriting

Menlo partner Tim Tully described agentic AI as a new infrastructure category that needs performance, efficiency, and control. John Vrionis of Unusual Ventures focused on the challenge of giving AI systems meaningful access to valuable enterprise data without sacrificing reliability or trust. Both views point toward the same market handoff: model capability is becoming useful enough that the infrastructure around access and execution can determine whether an enterprise deploys it.

That thesis is visible elsewhere in the market. DevCuration's coverage of Arcade.dev's authorization layer shows investors funding the controls that let agents act across enterprise systems. Its reporting on Baseten's AI inference infrastructure reflects the parallel demand for systems that make production AI faster and economically sustainable.

OliverAI is trying to combine several of those pressures around the analytical data layer. The company still has to show that buyers want one architecture spanning query performance, compute efficiency, model coordination, and policy enforcement, rather than separate products for each problem. The financing gives the founders time to put that proposition in front of enterprise workloads.

What OliverAI Must Prove Next

No named customer, revenue figure, independent benchmark, formal certification, or production-adoption metric was disclosed with the round. That leaves the next evidence straightforward: customer-specific results that hold under real schemas, concurrency, security requirements, and failure conditions. The strongest proof would connect query speed and compute reduction to a governed agent workflow that an enterprise trusts enough to keep running.

OliverDB does not need every data warehouse to disappear for that evidence to matter. It needs to show that an agent-ready layer can improve the economics and control of work already pressing against existing infrastructure. Every evaluation now carries the same question the investors accepted: can a database built for machine users earn an architecture decision from the people responsible for the cost, security, and consequences?