My project Called CrashSight

CrashSight — Autonomous Incident Triage & Self-Healing Agent
Multimodal in-situ crash diagnosis, one-click safe client remediation, and automated repository escalation powered by Gemma 4 and Gemini.

⚡ Overview
Traditional crash reporting tools (Apple Crash Reporter, Windows Error Reporting, and Crashlytics) act as one-way telemetry systems: they collect raw crash data for future developer sprints while leaving the end user stranded with a broken application.

CrashSight inverts this paradigm:

Edge Self-Healing First: Diagnoses root causes using visual screenshots and symptoms, then generates and executes safe local remediation commands, such as cache clearing, lockfile cleanup, and GPU flag toggling, to restore functionality immediately.
Dual-Engine Inference: Supports fully air-gapped local inference via Gemma 4 (Ollama) or cloud-based multimodal analysis via the Gemini API.
Autonomous Escalation: When a crash originates from an upstream code defect that cannot be fixed locally, CrashSight prepares a reproduction-ready issue for submission to GitHub or the application's official support desk.
🧭 Pre-Flight Checklist (Cold-Restart Guide)
Follow these steps whenever starting CrashSight for a live demo.

1. Start Ollama (Local Gemma 4) If you are using local inference, start the Gemma 4 model:

ollama run gemma4
If the model is not installed, pull it first:

ollama pull gemma4

1. Free Port 8000 If the backend fails with [Errno 48] Address already in use, stop the process occupying port 8000.

macOS / Linux:

lsof -ti :8000 | xargs kill -9 2>/dev/null || true
Note: This command forcefully terminates processes using port 8000. Make sure the process is safe to stop before running it.

1. Verify API Keys Check that the backend/.env file contains the required environment variables:

GEMINI_API_KEY=your_gemini_api_key_here
GITHUB_TOKEN=your_github_personal_access_token_here
Ensure the credentials are valid and have the permissions required for the features you intend to use.

1. Launch the Backend from a Native Terminal On macOS, launch Uvicorn from the native Terminal.app if VS Code's integrated terminal triggers macOS privacy permission prompts when executing remediation commands.

Grant only the permissions necessary for the application's intended operations.