CaseMate – document intelligence for litigation
Automated document management. Made simple.
CaseMate turns a case full of scans into a record you can search. Claude, ChatGPT or Gemini connects to it and works the file where it lives.
Create a case. Import a folder.
Name the case, point CaseMate at a folder of scanned discovery, and it reads every page into searchable text. The recording shows it from a fresh install.
Transcript
A screen recording of CaseMate, from a fresh install to a read production, made in the product itself working one fictional matter. There is no audio; everything on screen is written out below.
What happens
- The recording opens on a fresh install: CaseMate’s header, the Cases page, one “Open a new case” form, and the words “No cases yet.” beneath it. Nothing has been created and nothing has been imported.
- The pointer clicks into the Case name field and types “Rowan v. Delmar Freight Systems”, and under the field CaseMate answers with the id it will use — “Will be created as ‘rowan-v-delmar-freight-systems’”.
- Create case is clicked. The form clears, and where “No cases yet.” stood there is now the case: a card reading “No documents yet.” and “Import a folder to bring this case’s documents in”, carrying the three category badges every new case starts with — Evidence, Research, Precedent.
- The case name is clicked and its page opens: the caption as the heading, one “Add documents to this case” panel offering “Import a folder” and “or drop a folder here”, and the case’s still-empty tabs — Documents, History, Findings, Claims / Issues, Entities, Drafts.
- The pointer clicks Import a folder. The button turns to a spinner, a line under it reads “Waiting for you to choose a folder to import…”, and Windows’ own “Browse For Folder” dialog opens onto “Select a folder”.
- The folder tree is scrolled to “Production 003”, it is selected, and OK is clicked.
- CaseMate copies the six files into the case, and a green banner lands: “Imported 6 files — casemate is processing them”, with “Watch progress, or keep working — the batch runs on its own.”
- The pointer clicks Watch progress, and the batch’s live page opens without a reload, onto a run already counting.
- Left of the split is the page image being read. Coloured boxes land over one block of text after another — headings, production stamps, ruled rows, body paragraphs — as each block is located on the page.
- Right of the split, those same blocks arrive as lines of text in the live feed, in the order they were read.
- The counters climb through the production, each file closing as its last page lands, and the run finishes inside the recording — nine pages of nine, six files of six, none failed, marked Done, at 51.8 seconds elapsed. The recording holds on the finished state, then restarts.
What is on screen
- The matter
- Rowan v. Delmar Freight Systems, a fictional trucking case. No client material appears in any recording on this site.
- The fresh case
- Created in the recording as “rowan-v-delmar-freight-systems”, carrying the three categories every new case starts with: Evidence, Research, Precedent.
- The production
- “Production 003” — six scanned, image-only PDFs, nine pages in total.
- The documents
- 0001_Incident_Report.pdf (2 pages), 0002_Kessler_Deposition.pdf (2), 0003_Meridian_Regional_Records.pdf (1), 0004_Delmar_Correspondence.pdf (2), 0005_Rockvale_Maintenance.pdf (1), 0006_Supplement_C.pdf (1).
- The boxes
- Each coloured box is one block of text located on the page image. The line that appears beside it in the feed is that block’s text.
- The counters
- A status line counting pages and files, stat cards for files, pages, done, failed and elapsed, and a progress bar — ending at 9/9 pages · 6/6 files, none failed, in 51.8 seconds.
- Reading now
- The left panel is captioned “reading page N…” as it works through the production; the right panel is the live feed, the same page’s blocks filling in beside it in the colours of the boxes.
- Format
- 87 seconds, 1920×1152 pixels.
What CaseMate does
Every page becomes searchable text
Import a folder into the case and it is read page by page — image-only scans included.
See it in action
Ask in plain words. Land on the page.
One search matches meaning and exact wording across the whole case, and every result opens with the matched passage boxed on the scan.
See it in action
A production filed like a case file, not a folder
Categories, tags, and flags, with a filter for every one of them. What is not filed yet is marked, not hidden.
See it in action
Facts that carry their receipts
Every finding cites exact text on a specific page. Click it and you are looking at that page.
See it in action
Draft where the record lives
Word documents, versioned: every save lands a redline against the last, and the file exports as real .docx.
See it in action
Your AI works the case file itself
CaseMate is not the AI. Claude, ChatGPT, Gemini — any assistant that speaks MCP — connects to it, and every citation it hands back lands on a real page.
See it in action
On your own hardware. Never metered.
Case material leaves your machine only through the AI provider your firm has already approved. No per-page charges, no usage tiers, no AI bill beyond the subscription you already pay.
How CaseMate works with your AI
A production is read on your own machine, your AI answers a question with citations, and what it finds is filed back into the case as findings, issues, and claims.
Transcript
A drawn animation of the whole pipeline — not a screen recording. Nothing in it is the product itself; the recordings elsewhere on the site are. There is no audio; everything on screen is written out below.
What happens
- Scene one. A boundary is drawn and labelled YOUR MACHINE, and everything that follows happens inside it. A folder drops in — named “Production 003”, with “6 files” under the name — and spills six sheets, five of them scans and one a born-digital file. The caption reads: “A case shows up as boxes of scans. Depositions, productions, medical records — someone has to read every page.”
- Scene two. A CaseMate panel appears holding the case Rowan v. Delmar Freight Systems, and the sheets feed into it one at a time. Boxes pop over the page inside the panel and the matching text lands as lines in the feed beside it. A chip names which way each page went: “image-only — OCR, on this machine” for most of them, “text layer — read directly” for the one born-digital file, and “heavy pass” lighting once for a scan the quick read could not make out. A PAGES READ counter climbs from zero to nine as the bar fills. The caption reads: “CaseMate reads every page. Right here. Native text is read directly — image-only scans are recovered by OCR, on your own machine.”
- Scene three. The read pages become a lattice of dots labelled “the record, indexed — meaning and exact wording together”. A query types itself — “brakes below minimum” — and three page results land: the maintenance work order at 91% with its BELOW MINIMUM rows in bold, the ECM supplement at 74%, the collision report at 69%. The caption reads: “Indexed by meaning, not just words. Semantic and exact search together — and every hit is a page you can open, not a preview.”
- Scene four. A Claude Desktop window slides in beside the panel, joined to it by a wire labelled MCP and subtitled “the case file stays put”, with traffic running both ways along it. The question appears: “Was Unit 4180 actually roadworthy that night? Check the maintenance records.” A search crosses the wire, the index pulses, three result cards land in the chat, and a two-paragraph answer follows with a citation pill on each paragraph. The caption reads: “Your AI works the file where it lives. It searches the record from inside the chat — and answers with citations that land on pages.”
- Scene five. A CLAIMS / ISSUES panel fills with what the session left behind: two evidence cards and a precedent card, each carrying the production page it cites, then the issue they answer — “Was Unit 4180 roadworthy on March 14?” — and the claim above it, “Negligence — maintenance and roadworthiness”. The chat window slides away and the structure stays. The caption reads: “The chat ends. The case file keeps the work. Findings, citations, issues, claims — filed in the record, open for your whole team.”
- Scene six. The panel and the boundary fade back, and the animation closes on the CaseMate wordmark, the line “The case file your AI works on.”, and under it “Every page read. Every fact anchored. Nothing leaves your office.”
What is on screen
- What it is
- A drawn diagram of the pipeline, not a recording of the product.
- Six scenes
- The production arriving; the reading; the index and a search; the assistant over MCP; the structure that outlives the chat; the close.
- The matter
- The same fictional trucking case the recordings use, Rowan v. Delmar Freight Systems.
- The captions
- Five captions carry the animation, and every one of them is quoted in full above.
- Format
- 51 seconds, 1920×1080 pixels.
A closed beta starts at the end of August
A small group of litigators will run real matters through CaseMate and shape what ships.
Two worked examples, both from public records:
A 1,160-page presidential financial disclosure
Returned as a fully cited review in about forty minutes on a three-year-old laptop.
Read the review (PDF)The 2,926-page Kentucky v. Purdue Pharma court file
Returned as a page-cited trial notebook.
Read the notebook (PDF)Getting started is an installer and a folder of documents — no server, no IT project, no data migration.
Questions we get asked
Do my client files leave my office?
The reading, the searching, the tagging and the drafting all happen on your own computer. We never hold a copy of a privileged file, anywhere. Case material goes out only when you ask your AI a question about the case, and then it goes to the AI company your firm has already approved — no new company to vet.
Can the AI make things up?
AI can make things up. That is real, and anyone who tells you otherwise is selling something. What CaseMate does about it is refuse to let an answer float free: every fact comes back attached to a page. Click it and you are looking at the actual scan, the sentence it relied on boxed. You never have to decide whether to believe it — you look.
Is there a per-page charge?
No. Nothing per page, nothing per search — nothing is rented, so nothing is metered. The only AI bill is the subscription you already pay for Claude, ChatGPT or Gemini, and it does not go up because you asked four hundred questions instead of one.


