SnapTicket: Turn Client-Side Crashes Into Instant Engineering Tickets

#hf26challenge

🎫 SnapTicket Agent

Autonomous Incident Triage & Telemetry Fusion Engine

⚡ Executive Summary

SnapTicket bridges the critical chasm between chaotic client-side runtime crashes and structured engineering issue trackers. By fusing low-level macOS diagnostic telemetry (.ips reports) with visual error screenshots, SnapTicket leverages Gemma 4 to eliminate ambiguous bug reports, diagnose root causes in seconds, and dispatch production-ready tickets seamlessly.

🛑 01 // The Friction: The "It Crashed" Dilemma

When an application fails in production, teams face a massive information gap:

The User Side: End-user reports are notoriously vague ("the screen froze"), lacking essential stack traces, exception codes, and environment metadata.
The System Side: Native OS crash logs (.ips files) contain invaluable post-mortem data—memory registers, faulting thread dumps, and dyld images—but are buried in cryptic system directories (~/Library/Logs/DiagnosticReports/).
The Manual Bottleneck: Manually correlating when a user captured a visual glitch with which thread faulted in the system logs wastes hours of engineering triage time.

🧠 02 // The Solution: Autonomous Telemetry & Visual Fusion

SnapTicket automates this pipeline entirely through a local-first, multimodal approach:

┌────────────────────────┐      ┌──────────────────────────────────┐
│ Visual Glitch Capture  │  +   │ macOS Telemetry (.ips Log Trace) │
│ (Retina Screenshot)    │      │ (Faulting Thread, Exception Code)│
└───────────┬────────────┘      └────────────────┬─────────────────┘
            │                                    │
            └─────────────────┬──────────────────┘
                              ▼
        ┌──────────────────────────────────────────────┐
        │        Gemma 4 Multimodal Reasoning          │
        │  - Correlates UI error state with log dumps  │
        │  - Identifies target app category (Native/3rd)│
        │  - Formats reproduction steps & diagnostics  │
        │  - Dynamically deduces canonical portal URL  │
        └─────────────────────┬────────────────────────┘
                              ▼
        ┌──────────────────────────────────────────────┐
        │       Privacy & Consent Control Studio       │
        │   [Attach Visual Evidence] vs [Text Only]    │
        └─────────────────────┬────────────────────────┘
                              ▼
                 Automated Dispatch Pipeline
                 ├── Channel A: Authenticated GitHub REST API
                 └── Channel B: Human-Assisted Web Portal Co-Pilot

🛠️ 03 // Core Architecture & CapabilitiesPassive Telemetry SensingNon-intrusively scans ~/Library/Logs/DiagnosticReports/ for recent incident dumps, extracting exact exception codes and faulting process names instantly.Context-Aware ClassificationDeduces whether an app is a Native macOS App (routing toward Apple Feedback Assistant) or a 3rd-Party App (routing to respective repositories).Multimodal Inference (Gemma 4)Downscales Retina captures to optimize memory headroom, fusing visual pixel evidence with stack trace telemetry into a structured Markdown report.Binary Privacy Consent GateEnables total data custody with a binary toggle: attach visual proof or strip images completely for an air-gapped, text-only diagnostic dispatch.Dual-Channel DispatcherPushes clean Markdown tickets directly via authenticated GitHub REST APIs or stages them via a browser co-pilot paired with native pbcopy clipboard fallback.📊 04 // Technical SpecificationsComponentTechnology / ImplementationRoleCore AI EngineGemma 4 (via local Ollama or Google AI Studio)Multimodal visual and log cross-reasoningBackend FrameworkFastAPI + Uvicorn (Async Python)Non-blocking REST API endpointsOS Telemetry ParserCustom Python log sensor (core/log_sensor.py)Extracts .ips exception vectors & threadsFrontend UIVanilla HTML5 / CSS3 (Zero-dependency)Responsive dashboard with animated progress trackingDispatch PipelineGitHub REST API & Playwright / pbcopyAutomated issue creation & browser autofill🚀 05 // Quickstart GuideGet the project running locally in seconds:Bash# 1. Clone the repository
git clone https://github.com/AtharvVhanalkar/snapticket-agent.git
cd snapticket-agent

2. Install dependencies

pip3 install -r requirements.txt

3. Configure environment

cp .env.example .env

4. Launch the agent server

python3 main.py
Open http://127.0.0.1:8000 in your browser to start triaging.🔒 06 // Security & Data SovereigntyLocal Sovereignty: When running in local mode via Ollama, diagnostics, logs, and screenshots never leave your machine.Credential Protection: API keys and tokens are restricted strictly to local .env variables and excluded from version control.

DEV Community