Home / Platform
Detect. Review. Redact. Prove.
Most redaction software stops at the third step. The fourth is the one that decides whether a release survives an appeal, and it is where Redactor is built differently.
The Four Stages
1. Detect
The file is analyzed for the object and PII classes you selected. Detections are stored as timed data, indexed with their positions over time, so the studio can show where each object appears across the timeline rather than frame by frame.
Detection runs at upload, or on demand against something already in the library. Scope is selectable by duration and by class, so part of a video processes on its own.
2. Review
Nothing is released on the detector's say-so. An operator merges duplicate tracks, adjusts boxes, deletes false positives, and draws what the detector missed. Redactions organise into named layers, so one asset serves several disclosure tiers without maintaining separate copies that drift.
Automated detection proposes; a person disposes. That step is what makes the output defensible.
3. Redact
The redaction is burned out of a new rendition, not layered over the original. What happens to the source is a configured policy with three options, not a side effect.
For volume, the same pipeline runs with no operator opening the file, permissioned per format, driven by reusable templates for recurring document types.
Permanent redaction · Automation and scale
4. Prove
Every redaction carries the exemption it was made under, printed onto the redaction itself and into the transcript. Coverage reporting states what share of a framework's detection set was actually caught. The custody trail records every action with user, email, IP and timestamp.
This is the half of the job the category mostly ignores.
Across Every Format
| Video | Faces, persons, plates, vehicles, weapons, screens, tattoos, on-screen text, custom objects, tracked across frames |
| Audio | Spoken PII found by reading the transcript rather than listening, with speaker separation |
| Documents | Text, objects embedded inside PDFs, tables by row and column, non-Latin scripts |
| Images | 187 image formats out of 315 total, identity documents, 360-degree panoramas |
| DICOM | Medical imaging, pixels and header tags as separate operations |
Supporting capabilities: OCR and text · File formats · API
Where It Runs
The same engine runs in shared SaaS, dedicated SaaS, private cloud, on-premises, hybrid and air-gapped environments. AI inference is self-hostable, so detection need not call out of your network.
See It on Your Own Files
Send us something difficult and we will run all four stages on it.