Skip to content

Every document, every person, every claim — one connected case file.

CrimeCollab turns public police, court, and FOIA records into one searchable graph of documents, people, locations, vehicles, and evidence. Upload a file and it’s read, extracted, and cited automatically — and when another investigator’s work touches yours, you can adopt it one fact at a time.

Free to join Cases are private by default Built on already-public records

Illustrative case. All names, documents, vehicles, and locations shown are fictional.

The model

You’re never starting the case from zero.

Today

The documents live in three Drive folders and a forum thread. The same person’s name is spelled four ways. Nobody can tell which claims are sourced and which have just been repeated a hundred times.

On CrimeCollab

One corpus, one graph. Every person, place, and vehicle resolves to a single entity. Every claim carries the count of sources behind it, and the count of people who merely asserted it.

One graph, not one folder

Documents, people, organizations, locations, vehicles, evidence, and events all live on the same structure. Connecting two of them connects them everywhere.

Every view is a lens

The timeline, the case map, the directory, and the narrative are not separate databases. They are four ways of looking at the same case.

Work arrives as hints

When another member’s research touches yours, you get a leaf. You adopt facts one at a time, onto your own copy — never a bulk overwrite.

How it works

Three steps from a folder of PDFs to a working case.

  1. Bring the records in

    Drag in police reports, court filings, FOIA releases, and transcripts — one at a time or in bulk from a shared Drive folder.

  2. Let it read them

    Each file is OCR’d, run through automated sensitive-data masking, minted as a verified record, and mined for every person, place, and vehicle it mentions.

  3. Investigate across the graph

    Search it, map it, put it on a timeline, and ask it questions — with a citation on every answer.

Capabilities

Everything the case needs, on one graph.

Nine tools that all read and write the same connected case file, so work done in one shows up in the others.

Ask the case a question. Get an answer with citations.

Hercules answers only from the documents in the case and links every claim back to the page it came from. A mode switch keeps what the evidence supports visually separate from what is still a theory — and when it wants to add a fact or an event, it proposes and you accept.

Drop in a PDF. Get a searchable, connected record.

Upload a document and it is OCR’d, masked, verified, and mined for every person, place, and vehicle it names — automatically.

The corkboard, minus the corkboard.

Pan and zoom a board of people, documents, and events joined by string, and start from a board suggested by what the case already knows.

A timeline that admits what it doesn’t know.

Events group by month and day, uncertain dates render as ranges instead of false precision, and long empty stretches collapse so you can see the shape of the case.

One search across every case.

⌘K searches documents, people, organizations, locations, vehicles, and evidence at once — and it crosses case boundaries by design.

Read what the black bars hid.

CrimeCollab measures each redaction bar and reads any exposed initial, then ranks candidates from names that already appear in the case. Every fill-in is your own private interpretation: it is never published as fact, never shown to others as an assertion, and never alters the source document.

The whole case, readable in one sitting.

Read it as continuous prose with a Key Figures sidebar, or take the auto-generated broadsheet digest — useful when the deliverable is a script, not a database.

Places on a map, people in a directory.

Locations geocode onto a map, and people, organizations, vehicles, and evidence each get a directory that stays in sync with the documents.

A room for the work, and a record of it.

Threaded channels for discussion, and an activity feed showing every change made across the case, by whom, and when.

Method

Every claim shows its work.

Online case research goes wrong in one specific way: a rumor gets repeated until the repetition looks like corroboration. Every number on this page is designed to make that impossible to miss.

  • Two counts, never one. Each fact shows how many independent sources back it and, separately, how many people have asserted it. A claim repeated a hundred times never looks like a hundred sources.
  • The source bytes never change. The uploaded file is immutable. Everything built on top of it is a separate, revisable layer with full history and rollback, so you can always get back to the paper.
  • Evidence and speculation stay apart. The assistant draws a visible seam between what the documents establish and what is inference, and labels each side.
  • Consensus is an aggregate, not a verdict. Where independent investigators converge, you see the weighting and the counts behind it — not a single trust score standing in for judgment.

Who it’s for

Built for anyone who works from the record.

Researchers and case communities

Stop rebuilding what someone already built.

  • Inherit facts from other members’ work, one at a time
  • One shared graph instead of a dozen private spreadsheets
  • Channels and an activity feed so a group works in the same place
See the model

Podcasters, writers, and creators

Ship faster, with everything sourced.

  • Read the case as continuous prose, or as a printed-style digest
  • Every fact carries its citation, so the script has a footnote
  • Key figures, timelines, and photos already assembled
See the capabilities

Journalists and investigators

Discovery you can defend.

  • Search people, organizations, locations, and vehicles across cases
  • Explainable connection suggestions you agree to or skip
  • Provenance counts on every claim you might publish
See the method

Attorneys, law enforcement, and cold-case units

Structure a document set without losing the chain.

  • Immutable source artifacts with append-only revisions and rollback
  • Per-tool visibility, private by default, with named or group access
  • Automated sensitive-data masking on ingest
See how records are handled

Responsible use

Public records, handled carefully.

Public sources

The corpus is already-public police, court, and FOIA material. Source bytes are stored unmodified, with a full revision history over everything built on top of them.

Private by default

Every case, and every tool inside it, starts private to you. You grant access to named people or curated groups, per tool. Nothing is public unless you publish it.

Interpretation stays labeled

Speculation is visually separated from evidence, personal annotations stay personal, and every fact shows how many independent sources back it.

CrimeCollab is a research tool. It doesn’t publish accusations, and it doesn’t replace an investigation.

FAQ

Questions people ask first.

What is CrimeCollab, exactly?

It’s a workspace for building a case file out of public records. You upload documents, CrimeCollab reads and structures them, and everything inside — people, locations, vehicles, evidence, events — becomes a connected graph you can search, map, and discuss. It’s a research tool for organizing and sourcing what’s already public. It doesn’t publish accusations and it isn’t an investigation.

What can I upload, and what happens to it?

Police reports, court filings, FOIA releases, transcripts, and similar public records, as PDFs or scans — individually or in bulk from a shared Google Drive folder. Each file is OCR’d by Google Document AI, run through automated sensitive-data masking, and minted as a verified record. The original bytes are stored unchanged; everything built on top is a separate, revisable layer.

Are my cases private?

Yes, by default. Every case and every tool inside it starts private to you. You grant access deliberately, to named people or to curated groups, and visibility is set per tool rather than all-or-nothing. Nothing becomes public unless you publish it. Personal annotations stay personal even inside a case you share with others.

Do I need my own documents to start?

No. You can join a case that already exists and start reading, searching, and asking questions immediately — the corpus, the timeline, and the entity graph are already there. Most people begin by exploring an existing case and contributing a document or a correction later. Starting your own case from a single PDF also works.

Is the AI assistant just guessing?

No. Hercules answers only from the documents in the case and shows the citation for each claim, so you can open the page it came from. A mode switch separates what the evidence supports from what’s still a theory, and the colour of the panel changes with it. When it wants to add a fact or a timeline event, it proposes — you accept.

What is redaction analysis?

Scanned releases often black out names. CrimeCollab detects each redaction bar, measures it, reads any exposed initial, and ranks candidates from names that already appear in the case. A fill-in is your own private interpretation: it is never published as a fact, never shown to other members as an assertion, and never alters the source document.

What does it cost?

Joining is free, and there’s no card required. Paid plans will come later for heavier use, and existing members will get notice before anything changes. Separately, members receive a monthly allotment they can use to tip contributors whose work they relied on — that’s a way to credit people, not a fee you’re charged.

Start with one document.

Bring in a single report and watch it become people, places, dates, and citations you can search. Or join a case that’s already underway and start reading.

Create a free account

Free to join Cases are private by default