AI Chat Web — Conversational Copilot
Overview
PlantPulse AI Chat Web is a conversational AI copilot web application for manufacturing floor operators.
Ask a question in natural language, and the built-in agent (AgentOrchestrator + ToolRegistry) automatically selects and invokes the tools it needs to retrieve real-time data, run AI analysis, and search the knowledge base. The platform's integrated MCP (real-time data), TimeSeries-Insight (AI analysis), and LightRAG (document search) tools are all consolidated into a single chat interface.
With the 2026-07 revision, chat orchestration is consolidated into the copliot-web built-in agent. The separate Langflow engine has been removed, and LLM inference now goes through the LiteLLM (gpt-4o) gateway.
Key Features
Natural-language access to plant data
- Ask in natural Korean → the AI automatically selects the appropriate MCP server and tool
- Intuitive dialogue such as "Show me the injector temperature" or "Show me the Daejeon plant structure"
- Automatic conversion from equipment name (Korean) to system ID
SSE real-time streaming responses
- Real-time streaming responses based on Server-Sent Events
- Watch the AI's analysis unfold as it happens
- Charts and diagrams render immediately, even mid-stream
Rich visualization
| Visualization type | Supported charts |
|---|---|
| ECharts | Time series trends, gauges, bar/pie/radar/heatmap/Sankey/treemap |
| Mermaid | Process flow diagrams, state transitions, anomaly diagnosis trees |
| Markdown | Sensor data tables, equipment lists, alarm history |
Multi-application management
- Manage multiple AI applications (apps) by purpose (general-purpose copilot, equipment diagnostics, document search, and so on)
- Switch system prompts and tool configurations per app
- Register, edit, and delete apps from the admin panel
Built-in agent tool routing
- The LLM (gpt-4o) analyzes the intent behind the user's question and selects the optimal tool automatically
- For compound questions, an agent loop invokes multiple tools in sequence (tool-calling)
- Tools: built-in
rag_search(LightRAG) andts_get_asset_health(Insight), plus dynamic tools from the platform's integrated MCP - The work sequence can be reviewed via "View activity" (AgentTraceStore)
Screen Layout
| Screen | Description |
|---|---|
| Main chat | Real-time dialogue with the AI copilot, with inline chart/diagram rendering |
| Application selection | Switch between AI apps by purpose |
| Chat history | Search past conversations and resume them |
| Admin panel | App management, user management, prompt template management |
Technology Stack
| Component | Technology |
|---|---|
| Backend | Java 21, Spring Boot 3.4, Spring Security 6.x |
| Frontend | JSP, Bootstrap, jQuery, ECharts, Mermaid |
| AI backend | Built-in agent (AgentOrchestrator + ToolRegistry) → LiteLLM (gpt-4o), SSE streaming |
| Database | PostgreSQL (metadata and chat history), Redis (cache) |
| File storage | MinIO (S3-compatible object storage) |
| Deployment | WAR (servlet container) or embedded Tomcat |
Main Settings
Configuration file: src/main/resources/application.properties (values can be overridden with environment variables)
| Item | Description | Default | Environment variable |
|---|---|---|---|
openai.api.url | LLM (LiteLLM) gateway URL | http://127.0.0.1:4000 | CH_OPENAI_API_URL |
openai.model | Chat LLM model | gpt-4o | CH_OPENAI_MODEL |
mcp.api.url | Platform integrated MCP base URL | (platform server-web address) | CH_MCP_API_URL |
mcp.api.token | Integrated MCP api_key | (set during installation) | CH_MCP_API_TOKEN |
mcp.tools.enabled | Whether integrated MCP tools are used | true | CH_MCP_TOOLS_ENABLED |
mcp.timeseries.api.url | TimeSeries-Insight API URL | http://127.0.0.1:8970 | CH_MCP_TIMESERIES_API_URL |
rag.api.url | RAG API URL | http://127.0.0.1:7115 | CH_RAG_API_URL |
rag.api.key | RAG API key | (set during installation) | CH_RAG_API_KEY |
sse.stream.timeout | SSE streaming timeout | -1 (unlimited) | — |
app.request.limit | Concurrent request limit | 10 | — |
server.servlet.session.timeout | Session timeout | 480m (8 hours) | — |
A single chat call can reach up to three backends — real-time data from the platform's integrated MCP (mcp.api.url),
AI analysis from TimeSeries-Insight (mcp.timeseries.api.url), and document search from the RAG API (rag.api.url).
If an answer comes back as "data cannot be retrieved," start by checking which of the three is down.
Characteristics
Web interface ready for immediate deployment
- No separate client installation — access with a web browser alone
- Responsive support for mobile and tablet
- Authentication and authorization based on Spring Security
UX optimized for the manufacturing floor
- Korean-native — natural understanding of plant terminology and equipment names
- Visual status indication (normal / warning / critical)
- Compound analysis results automatically formatted into structured reports
Enterprise-ready
- WAR deployment integrates with existing WAS environments (Tomcat, WildFly, and so on)
- JDBC-based user authentication — integrates with an existing user database
- Full on-premises support — no cloud dependency