Making detection faster: Fast mode, hardware, and smart defaults
Stript offers two detection depths: Thorough checks every finding in context; Fast skips the deepest analysis for a much quicker first pass. When to use which, how hardware tiers affect speed, and how to keep quality where it matters.
Updated
Everything in Stript runs on your own machine; that is the point. It also means speed depends on your hardware and on how deeply you ask Stript to analyze. Both are under your control. This guide explains the levers, what each one trades, and sensible defaults for common situations.
The biggest lever: detection depth
Stript has two depths, switchable in the settings and storable per detection profile:
- Thorough (default): every candidate finding is additionally checked in its sentence context. This is what tells a client named in a letter apart from the court deciding the case, and it keeps false alarms low. It is also the most compute-intensive step of the analysis.
- Fast: skips that deepest verification step. Detection is much quicker, and deliberately more cautious: more items are marked for your review rather than silently resolved, so nothing is lost; you just decide more yourself.
The honest framing: Fast trades your review time for machine time. On a modern machine, Thorough is usually right; on a slower CPU, with very long documents, or when you’re triaging a stack of files to find the relevant ones, Fast is the better default. You can always re-run a document Thorough before it leaves the house.
The second lever: hardware tier
On first launch Stript detects your hardware and picks a compute profile: Light (8 GB), Standard (16 GB), Performance (32 GB), Ultra (48 GB+). Higher tiers use larger, more accurate local models; you can override the choice in the settings. Two practical consequences:
- On an 8 GB machine, Stript runs a lighter mode: fully functional, but detection is slower and slightly less accurate than on 16 GB or more. If you’re buying hardware for daily anonymization work, 16 GB is the tier where Stript gets noticeably quicker and stronger; a GPU (Apple Silicon or NVIDIA) accelerates it further.
- The first run after installation downloads the AI models once; after that Stript works fully offline and startup is fast.
What else affects run time
- Document size and type: long documents and scanned PDFs (which are read on-device via OCR) simply contain more to analyze than a one-page letter.
- You don’t have to wait idle: results stream in as the analysis progresses, and you can start reviewing the first findings while the deeper verification is still running.
- Sensitivity and type toggles don’t change compute time much, but they change your time: disabling types you never anonymize (say, URLs in a matter where they’re irrelevant) shrinks the review list.
Recommended setups
| Situation | Depth | Notes |
|---|---|---|
| Outbound documents, client data | Thorough | Quality where it matters; the default. |
| Triaging many documents | Fast | Find the relevant ones quickly, re-run the keepers Thorough. |
| Older laptop, 8 GB RAM | Fast for daily work | Thorough for the final pass on sensitive documents. |
| Recurring matter types | Save it as a profile | One profile with Fast for intake, one with Thorough for outbound. |
One thing no setting changes: nothing is uploaded, at any depth, on any tier. Speed tuning in Stript is always local tuning; you can verify that yourself.