Delphina raises $7.5M to scale its AI context system

Delphina raised $7.5 million on November 12, 2026, and used the announcement to launch Context Factory, bringing the company’s total funding to $15 million. Context Factory uses AI agents to build and maintain the business context behind analytics answers.

Delphina says the new financing will scale the product and expand its go-to-market and customer-engineering teams. The bet matters because the company says accurate AI analytics depends on knowing which tables are trusted, how metrics are defined, and what experienced analysts know but have not documented. Companies usually assemble that context by hand, Delphina says, which leaves it stale as the business changes.

Delphina describes Context Factory as a working layer for the assumptions that connect tables, metrics and decisions. Agents keep that layer aligned with company systems, while customers retain control over the underlying context and its approval process.

Who backed the $7.5 million round?

Radical Ventures and Costanoa co-led the financing. Tenacity Capital and angel investors also participated, including Fei-Fei Li, Nobel laureate Guido Imbens, and early employees of Anthropic and OpenAI.

Delphina made Context Factory available with the announcement. The stated use of proceeds covers product scale, go-to-market hiring and customer-engineering expansion. In plain terms, Delphina has both built a product and raised the money needed to sell it.

What problem does Context Factory solve?

Delphina’s premise is that an AI analytics system needs more than access to a company’s data. It also needs the rules and judgment that make the data usable, including trusted tables, metric definitions and undocumented analyst knowledge.

Context Factory assigns that maintenance work to agents. Delphina says they can assemble an initial baseline from existing company data and documents in hours. They identify and repair gaps as employees ask questions, then route consequential changes to domain experts for approval.

Delphina treats context as a changing system rather than a static glossary. A hand-built map can stop matching the business as definitions, data and decisions change, which is why the product assigns agents continuous maintenance rather than a one-time setup. That separation is the core of the design: agents keep context current, while experts decide which definitions matter.

Where does the context come from?

Delphina’s product overview says Context Factory reads dbt projects, dashboards, documents and query logs. It continues reading those sources as they change, which lets approved context follow modifications across the company’s systems.

Approved changes go into a versioned knowledge base that the customer owns. Evaluation tests can draw on trusted dashboards that employees have already verified, giving the system a checked set of reference material.

The combination gives agents both formal definitions and evidence of how people already use the data. Versioning records which changes received approval, while trusted dashboards give evaluation tests a human-checked reference. That matters because the knowledge base and regression tests give humans a way to constrain and check updates that agents carry out.

What happens when a metric definition is missing?

An illustrated workflow shows what happens when a question exposes a missing metric definition. The agent searches dbt, Looker and Notion, identifies the owner of the relevant data table and requests confirmation in Slack.

After approval, the workflow publishes the definition, regenerates the original answer and saves the fix as a regression test. Delphina says that test stops the approved logic from silently regressing when the agent handles later questions.

The example separates retrieval from authority. The agent gathers evidence from technical files and documents, but it does not treat a search result as approval for a missing business rule. After the Slack confirmation, the workflow turns the human decision into a published definition, a corrected answer and a durable test for future behavior. A human approves the business rule; the agent applies it and preserves it for the next run.

How does Delphina control enterprise access?

For enterprise deployment, Delphina says connections use per-user, read-only credentials. Each customer receives a separate data plane.

Python and DuckDB run in an isolated Firecracker microVM with no internet access. The system logs prompts, queries and knowledge references, and customers can export those records.

Under Delphina’s design, read-only credentials narrow what a connection can change before a query runs. The isolated microVM removes internet access while code runs, and the logs leave a record of prompts, queries and references for later inspection or export. Delphina says these controls make agent activity inspectable while restricting access to company systems and external destinations. The security model rests on narrow permissions, isolated execution and an exportable audit trail.

Who is using Delphina?

Delphina reports that business teams now ask 73% of the questions on its platform. CEOs are active users at more than half of its customers.

At Substack, more than 70% of employees use the product, including CEO Chris Best. Delphina also names LATAM Airlines, Upstart, Udemy and Coursera as customers.

Best described the effect on Substack’s decision-making: “Now our teams get accurate answers in seconds instead of waiting weeks in a queue. It’s changed how we make decisions.” The account makes the claimed time saving concrete.

Delphina provides the usage figures itself, but they show where the company expects the product to sit. Business teams already account for most questions, and CEO participation extends beyond the people who build data models. Chris Best’s account gives the adoption claim another measure: answers in seconds instead of a queue lasting weeks. The company wants analytics to become a routine business tool rather than a service limited to data specialists.

Who runs Delphina?

CEO and cofounder Jeremy Hermann previously built and led Uber’s Michelangelo machine-learning platform. He also cofounded Tecton, and Databricks acquired that company.

President and cofounder Duncan Gilchrist led global pricing data science at Uber. Gilchrist said operations, sales, marketing and finance teams are now Delphina’s heaviest users.

The founders bring experience across machine-learning infrastructure, data platforms and business pricing. That experience fits Delphina’s attempt to sell context maintenance to both data teams and executives, rather than leaving AI analytics inside a specialist workflow. Their backgrounds match the company’s push into everyday business decisions.

Can trusted analytics scale beyond data teams?

The funding gives Delphina resources to expand the product and the teams selling and supporting it. The company frames one central test: whether adoption by business users and CEOs can grow without weakening expert approvals, versioned definitions and regression tests.

Those controls underpin the company’s promise of trusted analytics. The next stage will test whether agents can keep context current while more employees rely on the answers and customer-engineering support expands.

If Delphina preserves that governance as usage grows, broader adoption will strengthen the case for agents that maintain business context rather than merely generate answers. The financing starts that test with $7.5 million and a product already in market.