Deterministic local AI file organizer powered by the Laya non-autoregressive decision engine. Categorizes files in sub-35ms forward passes on local CPU without cloud token costs, network overhead, or telemetry leakage.
| ENGINE / METHOD | LATENCY / FILE | NETWORK CALLS | COST / 10K FILES | RAM CONSUMED | PRIVACY / LEAKAGE |
|---|---|---|---|---|---|
| Laya Decision Engine | < 35 ms | 0 (Air-gapped) | $0.00 | 18.4 MB | 100% Local (0 bytes uploaded) |
| Cloud LLM (GPT-4o-mini) | 850 ms - 1,400 ms | 10,000 HTTPS POST | $1.50 - $4.00 | ~30 MB (Node/Py) | High (Filenames & snippets sent) |
| Local 8B LLM (Ollama) | 350 ms - 900 ms | 0 (Local) | $0.00 | 6,400 MB (GPU/RAM) | Local (Heavy compute penalty) |
| Naive File Extension Regex | < 1 ms | 0 | $0.00 | < 2 MB | High Error Rate (Misclassifies .txt, .pdf) |
python3 laya_organizer.py ~/Downloads --dry-run
⢠1KB Text Ceiling: Only the first 1,000 characters of UTF-8 text are inspected. This guarantees O(1) constant-time extraction without reading gigabytes of log files into memory.
⢠Non-Destructive Collision Protection: If a destination folder already contains a file with the identical name, Laya automatically increments a _dup1, _dup2 suffix rather than overwriting existing records.
⢠Safe Quarantine: Any file scoring below the confidence threshold (< 0.50 by default) is moved to Uncertain/ for human inspection rather than misfiled.