In high-stakes e-commerce environments, every second of downtime translates to financial loss. The SRE Incident Intelligence Agent is an AI-native solution designed to automate the critical first response of an incident. It doesn't just report issues; it analyzes them multimodally (logs, screenshots, metrics), retrieves knowledge from runbooks, and executes autonomous remediation actions to lower the Mean Time to Recovery (MTTR).
Traditional incident management relies on manual triage, leading to:
- Delayed response times during off-hours.
- Human error in diagnosing complex microservice failures.
- Disconnected silos between monitoring tools and documentation.
A stateful AI agent that acts as a Level 3 SRE, capable of:
- Multimodal Ingestion: Analyzing text descriptions and error screenshots.
- Contextual Reasoning: Correlating APM metrics with historical runbooks via RAG.
- Agentic Action: Performing self-healing tasks (e.g., restarting pods) before escalating to humans.
The system follows a modular, event-driven architecture orchestrated by LangGraph.
- Ingestion Layer: FastAPI endpoints receiving multimodal webhooks.
- Brain (LangGraph): A directed acyclic graph (DAG) that manages the ReAct cycle (Reasoning + Action).
- Perception (GPT-4o-mini): Multimodal LLM that "sees" error screenshots and "reads" logs.
- Memory (ChromaDB): Local Vector DB providing RAG capabilities for runbook retrieval.
- Action Layer: Custom toolset for Jira, Slack, and Kubernetes remediation.
- Multimodal Analysis: Upload a screenshot of a terminal error or a trace; the agent identifies the failure signature.
- Autonomous Remediation: Low-risk self-healing actions are executed agentically to stabilize the system.
- End-to-End Traceability: 100% of agent thoughts and tool calls are traced via LangSmith.
- Closed-Loop Notifications: Automatic updates to technical teams (Slack/Jira) and original reporters (Email).
- Frameworks: LangGraph, LangChain, FastAPI, Streamlit.
- LLM:
gpt-4o-mini(via OpenRouter). - Data & RAG: ChromaDB, HuggingFace Embeddings.
- Observability: LangSmith, OpenInference.
- Infrastructure: Docker Compose, Python 3.11.
Detailed instructions are available in QUICKGUIDE.md.
- Clone the repository:
git clone https://github.com/faiber1986/SRE-Triage-Agent.git
cd sre-triage-agent- Setup your
.envfile withOPENROUTER_API_KEY,LANGSMITH_API_KEYandLANGSMITH_TRACING_V2=true. - Run the stack:
docker compose up --build
- Access the UI:
- Streamlit UI: http://localhost:8501
- API Docs: http://localhost:8000/docs
- LangSmith Trace: http://localhost:8000/langsmith
- AGENTS_USE.md: Agent use cases, safety guardrails, and observability evidence..
- QUICKGUIDE.md: Detailed step-by-step instructions for testing the end-to-end flow.
- SCALING.md: Technical assumptions, scaling strategy, and financial/risk assessment.
This project is licensed under the MIT License - see the LICENSE file for details.