GUT stands for Guided Unconscious Thinking — the fast, instinctive judgment layer (System 1) that acts before slow reasoning is needed. GUT is a fine-tuned Laya (the open 421M decision model by Convai Innovations): the judgment of our cloud decision engine, distilled into a local model that runs on your hardware. Classify anything. Navigate real websites. Keep every byte private.
Real footage: GUT steering a live browser through the ASIC business-name registration workflow — every choice, confidence and saved training row visible. It stops at the login boundary by design: the machine prepares, the human signs.
SEO pages, OSINT evidence, RF signals, monitor alerts, community threads — graded locally with calibrated probability distributions.
Operation + target decisions per click. Walked the ASIC registration workflow end-to-end; stops at credential walls on purpose.
The pipeline ships as an app: point it at your systems' recurring decisions and distill your own decision engine — workstation CPU or free Kaggle GPUs.
One architecture, ten specialists — served per-domain behind one local endpoint (POST /v1/systemone, the same wire protocol as our cloud engine).
| Domain | What it decides | Where it's used |
|---|---|---|
| monitor-triage | severity · wake-the-owner · cause | qalarc monitor |
| osint-grading | verified / likely / weak / junk | OSINT hub |
| gmux-routing | complexity · risk · tier routing | gmux agent fleet · gmux.ai |
| rfai-feed | RF transmission class + interest | RFAI monitor |
| rfai-intent | radio call intent + urgency | RFAI monitor |
| geo-attribution | AI-engine citation correctness | GEO pipeline |
| geo-audit-triage | finding priority p0-p2 | universal GEO analyzer |
| seo-grading | quality · intent · cannibalisation | SEO audits |
| compliance-marks | keep · canonical · on-theme | doof.ing library |
| chanalyse | thread topic (71) + safety | chanalyse monitor · 4chan analytics |
| browser-ops | operation + target per click · completion | autonomous navigation — filmed |
Every checkpoint faces a hard gate before serving: agree with the teacher ≥0.85 and keep calibration error ≤0.10 on held-out data. Nine training runs are in the audit trail; none has passed yet — and that's the system working. Best calibration so far: ECE 0.07 (chanalyse). No uncalibrated model ever serves.
PolyForm Noncommercial (code) · CC BY-NC 4.0 (datasets and weights). Commercial licensing via qalarc.