Product / Evidence intake

Drop in the discovery. All of it.

Documents, scans, photos, body camera, jail calls, interview audio. Every file is processed on your own hardware into searchable, citable text, and the original is preserved untouched.

State v. Bellamy · EvidenceSample data
FileTypeSourceStatusIntegrity
Incident report 25-04471.pdfPDF · 14 ppProsecutionReadysha256 9f3c…e21a
Search warrant + affidavit.pdfScanned PDF · 6 ppProsecutionReady · OCRsha256 41b7…08d4
BWC_Ferreira_2025-03-14.mp4Video · 42:18ProsecutionTranscribed · 37 scenessha256 c0e5…77f0
Interview_Okonkwo.m4aAudio · 18:05Prosecution
sha256 d8a1…3b9c
Scene photos (23).zipImages · 23ProsecutionDescribed · OCRsha256 5e77…a4c2
Phone extraction summary.pdfPDF · 212 ppProsecutionReadysha256 2ab0…f61e
Client timeline notes.docxWord · 3 ppDefenseReadysha256 b3d9…10c7
Incident report 25-04471 (copy).pdfPDF · 14 ppProsecutionDuplicate of #1sha256 9f3c…e21a
📄

Documents

PDF, Word and plain text. Scanned and photographed pages are OCR'd; garbled PDFs with broken text layers are detected and re-read from the page image so nothing is silently lost.

🎥

Audio and video

Jail calls, interviews, body camera and dash camera are transcribed with timestamps. Video is split into scenes, and each scene is described so visual events are searchable too.

🖼

Images

Scene photos, screenshots and exhibits get a written description and any visible text is extracted, so a license plate in a photo is as findable as one in a report.

🏷

Provenance

Each item is labeled by who produced it, so an analysis can say “the State's report claims” rather than treating your own notes as the prosecution's evidence.

🧬

Duplicates

Files are content-addressed. The same report produced twice is stored once and flagged, and a near-duplicate with a page missing is called out rather than hidden.

🔏

Chain of custody

Originals are stored read-only with a SHA-256 hash on intake. Every derived transcript or OCR layer points back to the original it came from.

Runs in the background

Drag a folder in and keep working. A processing bar shows what is queued, what is running on which GPU, and what is done. You can ask questions of the files that are already finished while the rest catch up.

Large discovery productions are normal. A case with hundreds of files and hours of video is processed unattended, and the workspace shows exactly where each item is in the pipeline.

ProcessingSample data
Queued
3 files
Transcribing
Interview_Okonkwo.m4a · 11:12 / 18:05
Describing
BWC_Ferreira scene 31 / 37
Done today
41 files · 1,238 pages · 5h 54m media
← Product overviewNext: Case analyses →