a qalarc local AI system — live on this domain

GUT — Guided Unconscious
Thinking

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-parameter decision model: our cloud engine's judgment, distilled to run entirely on your hardware. Classify anything. Navigate real websites. Keep every byte private.

20ms
per decision · GPU
$0
marginal cost
100%
on your hardware
10
decision domains

Watch it decide

Real footage — no mockups. GUT steering a live browser through the ASIC business-name registration workflow: every option it scanned, the candidates it weighed, the choice it made, and the training row it saved. It stops at the login boundary by design.

Why GUT exists

Classification with general models is slow, hallucination-prone and hard to categorise. LLM agents struggle to navigate websites autonomously. And private data cannot be handed to an AI provider at all.

01

General models

Seconds per verdict. Confident hallucinations. A bill for every call. And a taxonomy you cannot control.

02

Browser agents

LLM agents misclick, loop and lose the plot — autonomous website navigation mostly fails in practice.

03

Private data

Data you cannot send to a provider never gets graded. Private stays useless — everywhere else.

The loop

The fleet teaches while it works. Three moves, repeated forever.

A

Collect

Every verdict our systems make — a grade, a verification, a route, a click — is logged as a training row with its full probability distribution. Soft labels keep the teacher's hedging.

B

Fine-tune + gate

RLCD reinforcement learning scored by proper statistical rules, then a hard gate on held-out data: match the teacher ≥0.85, keep calibration error ≤0.10. Pass or it never ships.

C

Serve

A passing checkpoint registers behind POST /v1/systemone — the same wire protocol as the cloud. One URL flips any app from cloud to local.

The ten decision domains

One architecture, ten specialists — each served per-domain behind one local endpoint. The systems they power are cross-linked below.

chanalyse threads
0
SEO gradings
0
RF signals
0
alerts triaged
0
OSINT verdicts
0
library marks
0
browser hops
0
DomainWhat it decidesWhere it's used
monitor-triageseverity · wake-the-owner · cause qalarc monitor
osint-gradingverified / likely / weak / junk OSINT hub
gmux-routingcomplexity · risk · reasoning · context → tier gmux agent fleetgmux.ai
rfai-feedRF transmission class + operator interest RFAI monitor
rfai-intentradio call intent + urgency RFAI monitor
geo-attributionAI-engine citation correctness · escalate GEO pipeline
geo-audit-triagefinding priority p0-p2 · fix-now universal GEO analyzer
seo-gradingquality · intent · cannibalisation · change significance universal GEO analyzer
compliance-markskeep · canonical · on-theme doof.ing library
chanalysethread topic (71) + content safety chanalyse monitor4chan analytics
browser-opsoperation + target per click · completion autonomous navigation — filmed above

Beyond the video

What the 80 seconds cannot show.

W1

One wire protocol

GUT serves the exact same API as the TypeSafe Jev cloud engine — typed questions (choice / score / noul) in, calibrated probabilities out. The X-Laya-Domain header picks the specialist. One URL flips cloud to local.

W2

Calibrated, or silent

Every checkpoint must agree with its teacher on held-out data AND keep its probabilities honest. Nine runs are in the audit trail; none has passed yet. The best calibration so far: ECE 0.07. Nothing uncalibrated ever serves.

W3

The data moat

Every training row comes from live qalarc operations — real monitors, real scans, real browser walks. Soft probability labels throughout. This dataset exists nowhere else and grows every day the fleet runs.

W4

Autonomy with boundaries

Task-directed navigation stops at credential walls by design. In the film above, GUT walked the ASIC registration flow and halted exactly at the login — the machine prepares every step, the human signs.

W5

System 1 / System 2

GUT is the reflex: pattern-matched verdicts in ~20ms. Novel situations escalate to big reasoning models or humans. Browsers get instincts; judgment stays where it belongs.

W6

Self-improving

Every cloud decision the fleet makes becomes training data for the local model that replaces it. The more the system runs, the less it needs the cloud.

System requirements

Modest by design — the whole point is running where your data lives.

UseMinimumComfortable
Serve a checkpoint8 GB RAM · any x86/ARM CPU (~2.5 s/decision)Any 4 GB GPU (17-23 ms/decision)
Fine-tune small domains16 GB RAM workstation, CPU-only (~1 h per 1k rows)One 16 GB GPU (T4 class) — minutes
No GPU at allThe pipeline pushes itself to Kaggle as a ready kernel — free 2×T4 sessions do the training
SoftwarePython 3.11+, PyTorch (CPU or CUDA), the laya package (Apache-2.0), Playwright for browser capture
Operating systemLinux tested (Ubuntu / Arch); macOS + Windows serve fine on PyTorch CPU/CUDA

Train your own

The pipeline ships as an app. If your systems make recurring decisions — grades, routes, labels, clicks — you can distill them into your own GUT in three moves.

1

Capture decisions

Point the collector at your workflow. Every verdict is logged as a training row with its full probability distribution. Works with any System One-compatible decider, human labels, or plain automation logs.

2

Fine-tune + gate

One command runs the RLCD loop and the calibration gate on held-out data. Workstation or free Kaggle GPUs. You receive a gate table — pass or fail — every single time.

3

Serve anywhere

A passing checkpoint registers into the local endpoint. Your apps talk to 127.0.0.1:8798 with the cloud's exact API — one URL flips between them. Nothing leaves your machine.

# 1 — capture: every Jev verdict becomes a soft-labeled training row
python scripts/jev_shadow_label.py --domain your-domain --n 1000

# 2 — train: RLCD loop + calibration + the gate (CPU here, Kaggle T4s there)
python mcp_finetune/train.py --dataset datasets/your-domain.jsonl --epochs 4 --device cpu

# 3 — read the gate table: agreement ≥ 0.85 and ECE ≤ 0.10, or it doesn't ship
cat checkpoints/your-domain-*/eval.json

# 4 — serve locally behind the same API as the cloud
curl 127.0.0.1:8798/v1/systemone -d '{"state":"…","questions":{…}}'
Get the full pipeline on GitHub

Free for non-commercial use

PolyForm Noncommercial 1.0.0 for code · CC BY-NC 4.0 for datasets and weights. Commercial licensing via qalarc — we make it easy.

GitHub — the pipeline qalarc.com More qalarc systems