You're an hour into a research session, you've found a 1910 census record that might be your great-grandmother's family, and you can't remember whether you've already seen it. The page name on Ancestry isn't ringing a bell. The notes you definitely took on this branch are somewhere — maybe in a Word doc, maybe in the family tree software's "notes" field, maybe in a Google Doc you started in 2022 and abandoned. You spend twenty minutes confirming you haven't reviewed this record, and another ten reorienting yourself to where you left off.

Most genealogy researchers run into this. The information accumulates faster than the system you use to hold it. Source citations live in the family tree software. Interview transcripts live in voice memos. PDFs of birth certificates live in a download folder. The actual *thinking* — the working theories, the brick walls, the leads you want to follow — lives nowhere stable.

A vault that holds the sources, the people, the brick walls, and your own working theories fixes most of it. The agent does the cross-referencing. (Family stories captured alongside formal records often live next to the kind of memory work in [Notes for Retirees: A Vault for the Years You've Got Plenty of Time For](/guides/personal-life/ai-notes-for-retirees/).)

## One parent page per family line, with the people underneath

In Docapybara, family-history research gets a parent page. Pages nest with no depth limit, OneNote-style, so a parent page called *Family history* can branch by line — *Maternal*, *Paternal* — and then by surname, and then by individual.

For most researchers, the right unit is the *individual* page. One page per ancestor of interest. The page can hold what you know, the sources you have, the open questions, and any photos or documents.

The agent treats the whole nested structure as one searchable pile. *"Find every person on these pages who lived in [town] in the 1880s."* The answer comes back with links to the pages and the relevant passages.

For researchers who already use family tree software (Ancestry, FamilySearch, MacFamilyTree, Gramps), the vault doesn't replace it — it complements it. The tree software does the relationship structure; the vault holds the research narrative, the sources in readable form, and the working theories. Both can point at each other.

## Sources that become readable, not just downloaded

The single most valuable habit in serious genealogy is full citation. Every fact has a source; every source is a real document with metadata.

PDFs of census pages, vital records, military files, immigration records — drop them on the relevant person's page, with a one-line citation at the top. PDFs upload and convert to markdown automatically, so the actual content of the document — names, dates, ages, places — becomes searchable text rather than locked inside a flat image.

This matters more than it sounds. *"What was [ancestor]'s occupation in the 1900 census versus the 1910 census?"* The answer comes back grounded in the actual records. *"Are there any places across these documents where [ancestor]'s name is spelled differently?"* The agent can pull the variants — useful when chasing the same person through poorly-indexed databases.

For records in foreign languages — German parish books, Italian civil registries, Yiddish death notices — the agent can translate as it reads. *"Translate the relevant passages from this German parish PDF and tell me what the entry on page three says about birth date and parents."* The answer is grounded in the actual scan.

## A people-database the agent can actually query

For larger research projects, an inline database of people becomes useful. The `:::database:::` directive lets a small spreadsheet live alongside whatever prose context you want.

Columns for name (with maiden name in parens), birth date, birth place, death date, death place, parents, spouse(s), and a link or note pointing to the longer page about them.

The agent can update it. *"Add the new ancestor I just confirmed — [name], born about 1843 in [town], died after 1910 in [other town], parents [X] and [Y]."* Row appears.

Once the database is populated, the agent can answer the questions that are normally tedious. *"List all the women in the database whose maiden name starts with M."* *"Pull every person in the database who was alive in [town] in 1885."* *"Find any pair of people whose dates overlap suggesting they might be the same person."* These are the cross-cuts that family tree software is bad at and that grind the research forward.

## A brick-walls page so the hard problems don't get forgotten

Every serious genealogy researcher has brick walls — the ancestors whose records simply can't be found, the conflicting documents that won't reconcile, the gaps the available archives can't fill. These problems get touched periodically and then forgotten between sessions.

A *Brick walls* page (or one child page per problem) keeps them visible. For each, the question, what you've tried, what you've ruled out, and what you'd try next. Three sentences, updated when there's progress.

The agent can re-orient you when you come back. *"It's been three months since I worked on the [name] brick wall — summarize the state, what I tried last, and what the most promising lead was when I stopped."* You re-enter the problem at the right place instead of re-reading everything you ever wrote about it.

For brick walls that involve cross-referencing across many sources — *"the [surname] family must have arrived in [port] between 1880 and 1885, but I can't find them"* — the agent can look across your existing material. *"Pull every reference in my vault to anyone arriving in [port] during that window, regardless of surname spelling, and flag any that look like plausible matches."* The candidate list narrows the search.

## Interviews with living relatives, recorded properly

The single most valuable source in family history is also the most perishable: the people still alive who remember things. Recording interviews with elders is the work that, once done, can never be redone.

Audio capture handles this. Drop a recording on a person's page (or on a dedicated *Interview with [name], date* page). You get a transcript with speaker labels — so when grandmother and aunt are both talking, you can see who said what.

The transcript is searchable like any other note. *"What did Grandma say about her father's brother in the interview last spring?"* The relevant passage comes back with a timestamp.

After the interview, ask the agent to mine it. *"Pull every name mentioned in this transcript that I don't already have a person page for, and list every place name."* You get a starting list of new individuals to research. *"Pull every story in this transcript that's worth turning into prose for the family book later."* The transcript becomes both a source and a structured set of leads.

For long, rambling interviews — the kind that wanders through eight decades of memory — the agent can summarize by theme. *"Summarize what this interview covers, grouped by family branch and rough chronology."* You get a map you can come back to without re-listening. (For the broader habit of catching memories in the moment, [The Capture Habit: Remembering the Things That Actually Matter](/guides/personal-life/capture-habit-remember-everything/) walks the practice.)

## A research log so the work compounds

The other low-glamour habit that pays off is keeping a research log. What you searched, where, what you found, what you ruled out. The reason it matters is that genealogy research is mostly negative results — a database that didn't have the record, a name that turned out not to be the right person, a hypothesis that didn't pan out. Without a log, you'll re-do the same negative searches.

A *Research log* page (or one entry per session) keeps it. *"Today: searched [database] for [name] +/- 5 years, found 3 candidates, all ruled out — see person page for reasoning. Searched [database] for parents — no results."* Two minutes at the end of a session.

The agent can pull the log when you're planning. *"What databases have I already searched for [ancestor], with what variations, and what's untried?"* The list comes back; you don't repeat work.

## A family stories page that's separate from the proof

Genealogy has two very different kinds of writing: the *proof* (with citations, careful reasoning, conservative claims) and the *story* (the narrative your family will read, with color, voice, and the stuff that doesn't quite have a citation but is part of who they were).

These benefit from being on separate pages. *Proof* on the individual's page — with sources. *Stories* on a dedicated child page — with whichever level of confidence is honest.

The agent can help with both, in different ways. For proof: *"Take what I have on [ancestor] and write a careful biographical paragraph — only claims supported by my sources, with citations."* For stories: *"Take Grandma's interview from last spring and pull the three best stories about her father's family into prose I could read aloud at the family reunion."* Two different jobs, two different drafts.

For researchers eventually compiling a family book, the stories page is where the book actually lives in draft form, building up across years of research. The agent can stitch it together when you're ready. *"Pull the stories pages across my paternal line into a single draft, ordered by generation, with a header for each individual."*

A complementary habit a lot of genealogists find helpful: a *Citation library* page with the citation format you've decided to use (Evidence Explained, Chicago, NGSQ — whichever), and the agent can format new citations against your standard. *"Format this source citation properly for the [census] type of record."* The agent uses your library as the standard. (For the broader history-of-the-family side once you've got the record straight, [Estate Planning and Will Preparation](/guides/personal-life/estate-planning-will-preparation/) covers the *Letters* habit researchers often layer in.)

## A starter shape that works on day one

If you're moving from "research scattered across software, downloads, and a few abandoned docs" to a vault, this is what we'd suggest:

- **A *Family history* parent page** — branch by line.
- **One page per ancestor of interest** — what you know, your sources (with PDFs), open questions.
- **A *People* database** — searchable, queryable, updated as you confirm people.
- **A *Brick walls* page** — keeps the hard problems visible.
- **A *Research log* page** — what you've searched and what you've ruled out.
- **An *Interviews* parent** — recordings + transcripts of every conversation with a living relative, dated.

That's it. No need for elaborate templates. The vault grows the way the research does, and after a year you'll have something a future descendant will be able to actually use.

The point isn't to make genealogy a database project. It's that the small amount of structure you keep means the next research session starts where the last one ended, the brick walls don't get forgotten between visits, and the stories your relatives told you survive in a form your grandchildren can read.

[Try Docapybara free](/accounts/signup/) — start with one ancestor's page, one source PDF, and a research log entry. The next time you sit down with the family tree, you'll know exactly where you left off.