Grafana AI Chat Assistant

An intelligent chat interface for Grafana that leverages Large Language Models (LLMs) to provide interactive assistance with monitoring, observability, and system analysis.

Quick Start

1. Install the plugin (see Installation)
2. Restart Grafana
3. Navigate to Apps > AI Chat Assistant
4. Configure LLM provider in Administration > Plugins > LLM App
5. Start a new chat conversation

Features

• Interactive AI Chat: Real-time streaming conversations with AI assistants
• LLM Integration: Works with OpenAI, Azure OpenAI, and other LLM providers via Grafana's LLM service
• MCP Tool Integration: Automatic tool calling with Model Context Protocol servers
  • Real-time tool execution tracking and status indicators
  • Support for multiple MCP servers
  • Ability to cancel long-running tool operations
• Multi-Session Management: Create, switch, rename, and delete chat sessions
• Backend Persistence: Database (PostgreSQL, SQLite) or file-based storage with per-user isolation and configurable limits
• Rich Message Display: Markdown rendering with syntax-highlighted code blocks
• Configurable Settings: System prompts, session limits, and MCP toggle

Screenshots

Prerequisites

• Grafana >= 10.4.0 with the LLM App plugin installed and configured
• LLM Service Configuration: API keys for OpenAI, Azure OpenAI, or other supported providers
• Optional: MCP Client plugin for enhanced tool capabilities

Installation

grafana-cli (Coming Soon)

Installation via grafana-cli will be available once published to the Grafana plugin catalog.

grafana-cli plugins install cisco-aichat-app

Manual Installation

1. Download the latest release from GitHub
2. Extract to your Grafana plugins directory
3. Restart Grafana

Docker/Kubernetes

Mount the plugin directory and configure in your deployment:

volumes:
  - ./plugins/cisco-aichat-app:/var/lib/grafana/plugins/cisco-aichat-app

environment:
  - GF_PLUGINS_ALLOW_LOADING_UNSIGNED_PLUGINS=cisco-aichat-app

Building from Source

Prerequisites:

• Go >= 1.24.1
• Node.js >= 18

git clone <repository-url>
cd cisco-aichat-app
npm install
npm run build
./scripts/build.sh

Configuration

LLM Service

Via Provisioning (Recommended):

# grafana/provisioning/plugins/apps.yaml
- type: grafana-llm-app
  orgId: 1
  enabled: true
  jsonData:
    openAIConfig:
      provider: "openai"
      model: "gpt-4"
  secureJsonData:
    openai_apiKey: "${OPENAI_API_KEY}"

Via UI:

1. Navigate to Administration > Plugins > LLM App
2. Configure provider settings and API key
3. Test connection

Chat Settings

Access via the gear icon in chat:

• System Prompt: Customize AI assistant behavior
• Max Sessions: Sessions per user (default: 10)
• Max Messages: Messages per session (default: 100)
• MCP Integration: Enable/disable tool calling

Development

# Development build with watch mode
npm run dev

# Production build
npm run build

# Run tests
npm run test

# Run linter
npm run lint

Backend Compilation

# Build for current platform
go build -o dist/gpx_cisco-aichat-app ./pkg

# Build for Linux/ARM64 (containers)
GOOS=linux GOARCH=arm64 go build -o dist/gpx_cisco-aichat-app ./pkg

See MULTIPLATFORM-BUILD.md for multi-platform options.

Architecture

• Frontend (TypeScript/React): Chat interface, session management, settings
• Backend (Go): REST API for persistence and session management
• LLM Integration: Via @grafana/llm library
• MCP Integration: Optional integration with MCP Client plugin

Backend Features

• Session lifecycle API (CRUD operations)
• Storage: PostgreSQL, SQLite, file-based, or in-memory (auto-detected from environment)
• Rate limiting (10 req/sec, burst 20)
• Input validation and XSS protection
• Health check endpoints

Troubleshooting

Chat not responding

• Verify LLM service is configured in Administration > Plugins > LLM App
• Check API keys are valid with sufficient quota
• Check browser console (F12) for errors

Plugin not visible

• Ensure plugin is installed and Grafana restarted
• Check Grafana logs for loading errors
• Verify plugin signing configuration

MCP tools not available

• Install and configure MCP Client plugin separately
• Ensure MCP servers are running and accessible
• Verify MCP servers are enabled in configuration

Security

• API Keys: Stored securely in Grafana configuration, supports environment variables
• Chat History: Per-user isolated storage with path traversal protection
• Access Control: Integrates with Grafana permissions
• Rate Limiting: Per-user limits prevent abuse
• Input Validation: XSS protection on all inputs
• Data Privacy: No data sent outside configured LLM provider

API Documentation

See docs/API.md for backend endpoint documentation.

Contributing

See CONTRIBUTING.md for development setup and guidelines.

License

Apache 2.0 License