Cisco AI Chat: An Intelligent Monitoring Assistant for Grafana

Cisco's open-source Grafana plugin AI Chat brings large language model (LLM) capabilities to the monitoring and observability domain. It supports multiple LLM providers, Model Context Protocol (MCP) tool integration, and multi-session management, helping operations teams efficiently handle massive monitoring data and complex system states.

Traditional monitoring dashboards are intuitive, but when facing complex issues, manual analysis of metrics, document review, or expert consultation is required. With the maturity of LLM technology, integrating AI dialogue and reasoning capabilities into monitoring tools has become a trend. Cisco AI Chat is a representative work of this trend, allowing operations teams to interact with monitoring systems via natural language to gain insights.

Multi-LLM Provider Support: Not tied to a single model; unified access to mainstream providers like OpenAI and Azure OpenAI via Grafana LLM services; MCP Tool Integration: Call external tools via Model Context Protocol to get real-time data (e.g., query metrics, retrieve logs), with tool status displayed in real time; Multi-Session Management: Create, switch, rename, or delete sessions; each session maintains independent context; Persistent Storage: Conversation history is stored in user-isolated files, supporting limits on the number of sessions and messages.

Architecture Design: Three-tier architecture (frontend for chat interface rendering, backend API for session/permission/rate limit handling, LLM integration layer for communication with external models); Security Considerations: Secure storage of API keys, XSS filtering for user input, protection against path traversal in conversation storage, rate limiting (10 requests/second by default, 20 burst); Deployment Methods: Supports installation via grafana-cli, manual extraction, Docker/Kubernetes integration, and provides local development guidelines.

Troubleshooting Assistance: Describe abnormal phenomena; AI guides users to view metrics, analyze causes, or call tools to get real-time data; Accelerated Newcomer Training: 24/7 online mentor that answers questions about system architecture, metric meanings, alarm processes, etc; Document Query Alternative: Get precise answers via natural language queries without manual searching through massive documents; Report Generation: Generate system status summaries and failure review reports based on conversation history.

Current Limitations: Primarily text-based conversations; chart interpretation and anomaly visualization capabilities need improvement; tool call accuracy and latency require optimization. Future Outlook: With the maturity of the MCP ecosystem and the popularization of multimodal models, it will support chart analysis, monitoring screenshot interpretation, visualization suggestion generation, etc.

Cisco AI Chat demonstrates how LLM capabilities can be seamlessly integrated into enterprise toolchains. It is a valuable open-source solution for AI-empowered operations and represents a new direction for the integration of monitoring tools and AI.