OnCall Agent Backend

This repository contains the backend service for an intelligent OnCall assistant. It provides conversational diagnosis, runbook retrieval, document indexing, and an AIOps workflow for incident analysis.

The frontend is intentionally not included. Any web or mobile interface can call the REST and SSE APIs exposed by this service.

Project Overview

The complete OnCall Agent project is designed as a human-AI collaborative incident response console. The backend provides the diagnosis and retrieval capabilities, while the frontend should focus on HCI-oriented interaction:

• transparent agent reasoning and diagnosis timelines
• voice-driven question input
• structured incident report visualization
• controllable automation levels
• command palette and slash commands
• searchable chat history and evidence highlighting

For the full frontend feature and API mapping table, see:

docs/frontend-interface-guide.md

Main Modules

src/main/java/org/example
├── controller      HTTP endpoints for chat, file upload, and health checks
├── service         Chat, AIOps, RAG, vector indexing, and vector search logic
├── agent/tool      Tools exposed to the agent runtime
├── client          Milvus client initialization
├── config          Spring Boot configuration classes
├── dto             Request and response objects
└── constant        Shared vector database constants

Features

• Multi-turn chat with session memory
• SSE streaming chat endpoint
• Runbook upload and automatic vector indexing
• RAG-based internal document retrieval
• AIOps diagnosis workflow based on planner, executor, and replanner roles
• Prometheus alert query tool
• CLS log query tool with mock mode for local demonstrations
• Milvus vector database integration

Requirements

• Java 17
• Maven 3.8+
• Docker Desktop
• DashScope API key

Configuration

Set the DashScope key before starting the service:

export DASHSCOPE_API_KEY=your-api-key

On Windows PowerShell:

$env:DASHSCOPE_API_KEY="your-api-key"

The default configuration in src/main/resources/application.yml enables mock data for Prometheus and CLS logs. This makes local classroom demos repeatable without requiring external monitoring systems.

For a real deployment, update:

• prometheus.base-url
• prometheus.mock-enabled=false
• cls.mock-enabled=false
• Spring AI MCP SSE endpoint for Tencent CLS

Run Locally

Start Milvus:

docker compose -f vector-database.yml up -d

Start the application:

mvn spring-boot:run

The API service listens on:

http://localhost:9900

The original Makefile also provides:

make init

This starts the vector database and uploads the sample runbooks in aiops-docs.

Core APIs

API	Method	Purpose	Frontend Usage
/api/chat	POST	Non-streaming chat response	Simple Q&A
/api/chat_stream	POST	SSE streaming chat response	Main chat workspace and agent timeline
/api/ai_ops	POST	SSE AIOps diagnosis report	Incident report dashboard
/api/upload	POST multipart/form-data	Upload runbooks and index them into the vector store	Knowledge base upload panel
/api/chat/clear	POST	Clear one chat session	/clear slash command
/api/chat/session/{sessionId}	GET	Read session metadata	Session status panel
/milvus/health	GET	Check vector database status	System health indicator

Chat

POST /api/chat
Content-Type: application/json

{
  "Id": "session-001",
  "Question": "How should I handle service unavailable alerts?"
}

Streaming Chat

POST /api/chat_stream
Content-Type: application/json

{
  "Id": "session-001",
  "Question": "Explain the current incident diagnosis steps."
}

AIOps Diagnosis

POST /api/ai_ops

Upload Runbook

curl -X POST http://localhost:9900/api/upload \
  -F "file=@aiops-docs/service_unavailable.md"

Milvus Health Check

GET /milvus/health

Notes for Course Demonstration

For an HCI project, this backend supports a human-AI collaborative OnCall interface. A separate frontend can visualize alert context, agent plans, retrieved evidence, and diagnosis timelines while calling these APIs.