Skillget

hck-GPT

agent
Security Audit
Warn
Health Warn
  • No license — Repository has no license file
  • Description — Repository has a description
  • Active repo — Last push 0 days ago
  • Low visibility — Only 5 GitHub stars
Code Pass
  • Code scan — Scanned 12 files during light audit, no dangerous patterns found
Permissions Pass
  • Permissions — No dangerous permissions requested

No AI report is available for this listing yet.

SUMMARY

Local AI diagnostic assistant for Windows PC monitoring. 76 intents, hybrid rule and LLM engine, bilingual PL/EN, context time-windowing, anti-hallucination fallback, and time-travel debugging. Built from real community requests. Part of PC_Workman HCK.

README.md

hck_GPT

AI diagnostic assistant embedded inside PC_Workman. Answers natural language questions about your PC in Polish and English using a hybrid rule+LLM engine.

Architecture

hck_gpt/
├── chat_handler.py          # Entry point — quick aliases, routing, help
├── panel.py                 # Chat UI panel, nav links, session display
├── insights.py              # InsightsEngine — habits, anomalies, teasers
├── process_library.py       # Process lookup from data/process_library.json
├── service_setup_wizard.py  # Windows service optimization wizard
├── services_manager.py      # Windows service stop/start manager
├── tooltip.py               # Process tooltip widget
│
├── intents/
│   ├── vocabulary.py        # 76 intents, PL+EN trigger patterns
│   ├── parser.py            # Intent parsing + confidence scoring
│   ├── lang_detect.py       # PL/EN auto-detection
│   └── ml_classifier.py     # Naive Bayes fallback classifier
│
├── engine/
│   └── hybrid_engine.py     # Routes intent → rule handler or Ollama LLM
│
├── responses/
│   └── builder.py           # All _resp_* handlers + MEGA features
│
├── context/
│   ├── system_context.py    # Builds PC context string for LLM
│   └── hardware_scanner.py  # WMI scan: CPU, GPU, RAM, disk, mobo
│
├── memory/
│   ├── session_memory.py    # In-session event log, spike tracker
│   ├── user_knowledge.py    # SQLite persistent user profile (AppData)
│   └── proactive_monitor.py # Background alerts (CPU/RAM/disk/uptime)
│
└── data/
    ├── live_sensors.py      # Real-time CPU/GPU/RAM snapshot
    └── metrics_store.py     # daily_summary() queries from hck_stats.db

Intent Coverage (76 total)

Category Intents
Hardware hw_cpu, hw_gpu, hw_ram, hw_storage, hw_all
Temperature temperature, throttle_check, gpu_temp_why
Performance performance, stats, perf_change, session_compare, pc_changes
Why why_slow, ram_why_high, processes, disk_usage_why
Diagnostics health_check, virus_check, disk_health, uptime, voltage_check
Gaming gaming_session, weekly_trends, fps_degradation, game_hardware_stress
New (1.7.5) fan_noise_history, driver_status, gaming_vs_work_time, process_identity, stale_apps, app_behavior_change, startup_slowdown, temp_comparison, crash_context, battery_drain_rate, power_after_restart
Optimization optimization, fan_speed, power_mode
Security process_info, security_check
Misc help, small_talk, unknown + 15 others

All intents have Polish and English trigger patterns. Language is auto-detected per message.

Key Features

Hybrid Engine

  • Rule path: known intents → deterministic _resp_* handler in builder.py
  • LLM path: open-ended or low-confidence → Ollama (local, no cloud)
  • Per-intent temperature and system prompt hint tuning

Context Time-Windowing

Each intent gets a history window matched to its nature:

"hw_cpu": 5,          # 5 minutes — live query
"health_check": 30,   # 30 minutes — recent session
"temp_comparison": 10080,  # 7 days — historical trend

build_llm_context_windowed(lang, minutes) builds the LLM context scoped to that window — tight windows strip stale patterns, wide windows append daily metric history.

No-AI-Slop Fallback

_no_data(intent, lang, what_missing) — returns a structured "data unavailable" response instead of fabricating an answer. Used when sensor data, history, or process lists are empty.

Time-Travel Debugging

_get_historical_comparison(metric, days, lang) — fetches live sensor value and compares to N-day average from metrics_store.daily_summary(). Returns formatted delta with direction arrow.

Micro-Benchmarking

_trigger_micro_benchmark(bench_type) — fires a background thread:

  • cpu_single: 1M sqrt operations, measures ops/sec
  • disk_seq: 32 MB sequential write+read, measures MB/s

Results stored in session_memory under micro_bench key.

Process Library

data/process_library.json — 241 processes with vendor, category, safety rating, typical CPU/RAM, and description. Used by process_identity and process tooltip widget.

Session Memory

Tracks per-session events, spikes, and response data. Later handlers can reference what was discussed earlier in the same session (discussed_this_session(), get_response_data(intent)).

Proactive Monitor

Background thread watching CPU, RAM, disk, and uptime. Fires non-intrusive alerts into the chat panel when thresholds are exceeded.

Usage

from hck_gpt.chat_handler import ChatHandler

handler = ChatHandler()
responses = handler.process_message("dlaczego mój komputer jest wolny?")
# returns list of formatted response strings (bilingual)

Quick aliases available in chat_handler.py — short Polish keywords map directly to intents without going through the parser (e.g. sterowniki → driver_status, bateria → battery_drain_rate).

Requirements

  • Python 3.9+
  • psutil — process and sensor data
  • wmi — hardware scanner (Windows only)
  • ollama — optional, for LLM fallback path (local install required)
  • sqlite3 — built-in, used by metrics_store and user_knowledge

Part of PC Workman HCK — github.com/HuckleR2003/PC_Workman_HCK
Developed by Marcin "HCK" Firmuga

Reviews (0)

No results found