ProvenRoots.ai: Proof-Grade Genealogy, Built and Operated by AI Agents

Mainstream genealogy products sell hints and tree matches: suggestions the user is supposed to verify and mostly never does. DNA tests sell a pie chart. Neither one can answer the question that actually matters to a family: can this lineage be proven?

Maxim Cloud Solutions built the system that answers it, and that system became a business.

The Problem

Lineage societies like the Daughters of the American Revolution, the Sons and Daughters of the Republic of Texas, and the Mayflower Society demand the opposite of hints. They require a documented chain where every parent-child link rests on a primary source, prepared on the society’s official form to their standard. Producing that is expert, tedious, correspondence-heavy work that professional genealogists charge thousands of dollars per line for.

The founding challenge was a family question that had gone unanswered for generations: could a lineage back to the era of the Republic of Texas actually be proven? Not guessed at. Proven.

The Idea

Use AI agents for the work no one can bill by the hour: sweeping digitized archives, reading 19th-century cursive, transcribing records, drafting correspondence, and assembling evidence packets, while keeping humans in charge of everything that matters.

The system enforces four human gates: evidence merge, data merge, outbound correspondence approval, and spending. No parent-child link enters the tree, and no letter leaves the firm, without a person approving it.

What We Built

• An agentic research pipeline that searches the open genealogical record, including FamilySearch’s official API, the Library of Congress’s Chronicling America newspaper archive, WikiTree, GovInfo Congressional records, the National Archives catalog, Texas state archives, and Dutch civil registers, through a shared, rate-limited, archive-respectful fetch layer. Agents machine-read historic handwriting and transcribe every captured record.
• A data-integrity engine: per-family data models compiled into GEDCOM, the genealogy interchange standard, through a corrections layer with safety assertions, merge guardrails that refuse to conflate same-named people, and a consistency gate for impossible dates, broken parent links, and orphans that the roughly 1,300-person unified tree passes with zero issues.
• A private family website built on Next.js 14, TypeScript, Tailwind, and Postgres with Drizzle, with self-hosted OIDC single sign-on and containerized deployment behind Cloudflare. Member accounts, person pages, family stories, and photos, with living-person privacy enforced at runtime per viewer, engineered in rather than bolted on.
• A closed research loop: when a family member submits a correction on the site, a background workflow automatically researches it across all sources and posts a fully cited evidence dossier for human review. The agent gathers and cites; a person decides.

The Outcome

The founding case answered the family’s question: two Sons of the Republic of Texas applications approved, with more staged and a Daughters of the Republic of Texas packet ready to file, built on primary manuscripts obtained by direct correspondence with county record offices.

The pipeline produced a four-line, roughly 1,300-person documented family tree with more than 560 source documents archived and fully transcribed, for roughly $170 in total records fees.

That engine is now the foundation of ProvenRoots.ai. Not found roots. Proven roots. It is launching with pilot families at founding pricing, with a professional line for heirship and lineage determination, serving title companies and estate attorneys, planned next.

For Maxim, ProvenRoots is what happens when agentic AI is pointed at work everyone assumed only patient experts could do: the machine does the sweep, the human makes the call, and the result holds up to the strictest documentary standard there is.