Qavren
← All products
Flagship/Genealogy · AI research

OmniSift

An AI research agent for the family story you've been meaning to write down.

Every family has a handful of stories nobody can verify and a box of documents nobody knows how to read together. OmniSift is the research agent that finally puts the box and the stories in the same place.

What it is

OmniSift is a multi-tenant SaaS for personal and family history research. You upload what you have — old PDFs, scanned letters, text-message exports, photos of certificates, links to obituaries — and OmniSift runs an AI agent that combines your private documents with active scraping of public web sources to build coherent profiles and timelines for the people you're researching.

It's built on ASP.NET Core 10, with a Blazor WebAssembly frontend, PostgreSQL + pgvector for semantic storage, and Microsoft Semantic Kernel orchestrating an agent with three core plugins:

  • Vector search over your uploaded corpus
  • Live web scraping for public records, news archives, obituaries, gravestones
  • Wayback Machine retrieval when the source you need has gone dark

Multi-tenant from day one — every record is tenant_id-scoped at the database level via PostgreSQL Row-Level Security. Your great-aunt's research stays in your tenant, full stop.

What it does well

Ingests what families actually have. PDFs of certificates, scanned letters, text-message exports (CSV/JSON), web pages. Each gets chunked (500 tokens, 50 overlap), embedded via OpenAI, and stored alongside everything else you've uploaded.

Agents the research, not just the search. Ask "what did Grandpa do between 1947 and 1953?" and the agent works the question: queries your uploads, scrapes a newspaper archive, pulls a census record, fetches a now-dead genealogy page from the Wayback Machine. It returns a synthesis with citations, not a list of hits.

Builds timelines, not just answers. OmniSift assembles a person's life as a structured timeline — events, places, relationships, sources — that you can read top-to-bottom and amend as new sources come in.

Cites everything. Every claim in a synthesized profile has a source link or document reference. Family history that can't be cited isn't history; it's a story. OmniSift treats the difference seriously.

Respects privilege. Tenant isolation is enforced at the database via RLS — not in application logic where a future bug can leak. The same isolation lets a single OmniSift instance serve unrelated families without bleed.

Who it's for

Family historians and genealogists who've outgrown spreadsheets, sticky notes, and Ancestry's UI but don't want to write SQL.

Estate attorneys and probate researchers who need to reconstruct a decedent's relationships and assets from a mix of personal and public sources.

Archivists and oral-history projects that need to synthesize across interview transcripts, scanned documents, and external records.

Anyone with a parent or grandparent whose stories you wish you'd written down before it was too late — and now you're working backwards from what's left.

What it looks like in practice

You upload 80 photos of letters your grandmother wrote between 1953 and 1988, a CSV of text messages with your mother covering five years of "remember when…" exchanges, and three obituaries from a town newspaper. You ask OmniSift to build a profile of your grandmother. Twenty minutes later you have a timeline with sources — places lived, jobs held, family events, names of friends she mentioned by first name only that the agent cross-referenced against the obituaries. You read it on a Saturday morning and learn three things you didn't know.

A probate researcher loads a decedent's box of papers, points OmniSift at three relevant newspaper archives, and gets back a draft asset/relationship map with cited sources, ready to hand to the estate attorney.

What it's not

It's not a public records database. OmniSift uses public sources but doesn't replace Ancestry, FamilySearch, or the U.S. census interface — it sits on top of them and synthesizes.

It's not a DNA service. Genetics aren't in scope.

It's not magic. If the source isn't in your uploads and isn't on the public web, OmniSift won't conjure it. It just makes sure nothing that is findable gets missed.

Status

Active development. Multi-tenant infrastructure shipping. Talk to us if your family has a box you've been avoiding.

Have a use case for OmniSift?

Talk to the people who built it.