🔍 RootSight

AI-Powered Incident Understanding & Action System

Recho "# RootSight" >> README.md

---

What It Does

Step	Manual (Today)	With RootSight
Timeline reconstruction	15–30 min across 5 tabs	Automated — structured timeline in seconds
Root cause analysis	Guesswork + pattern recognition	Evidence-backed hypotheses with confidence scores
Impact assessment	Incomplete mental estimates	Quantified blast radius and severity
Historical pattern matching	"I think we saw this before..."	Vector search across past incidents
Follow-up actions	Write Jira + Slack from scratch	Auto-drafted tickets and messages

Quick Start

Prerequisites

• Python 3.11+
• Google Gemini API key (Get one here)

Setup

# Clone and enter the project
cd RootSight

# Create virtual environment
python -m venv venv
venv\Scripts\activate  # Windows
# source venv/bin/activate  # macOS/Linux

# Install dependencies
pip install -r requirements.txt

# Configure environment
copy .env.example .env
# Edit .env and add your GEMINI_API_KEY

Run the Demo

# Build the memory index (first time only)
python scripts/build_memory_index.py

# Launch the Streamlit UI
python -m uvicorn app:app --port 8000

Run Tests

pytest tests/ -v

Architecture

RootSight uses a multi-agent pipeline where each agent owns one responsibility:

Alert → Trigger → Log → Timeline → RCA → Impact → Memory → Action → Brief

Each agent produces schema-validated output that flows to the next. If any agent fails, the pipeline continues with degraded confidence.

See architecture.md for full details.

Project Structure

src/
├── schemas/          # Pydantic data models
├── agents/           # Pipeline agents (trigger, log, timeline, rca, impact, memory, action, manager)
├── prompts/          # LLM prompt templates
├── integrations/     # External API clients (Gemini, Datadog, PagerDuty)
├── services/         # Business logic (normalizer, vector store, persistence)
└── utils/            # Helpers (logging)

data/                 # Seed data, mock responses, incident storage

License

MIT