-ops-agent ``` 2. Install the dependencies: ```bash pip install -r requirements.txt ``` 3. Install the required MCP servers: ```bash # Install the Kubernetes MCP server npm install -g @kubernetes-ai/mcp-server-kubernetes # Install the Prometheus MCP server pip install prometheus-mcp-server ``` > **Note**: Do not use the MCP servers located in the `deps/` directory for local development. These are customized versions: > - The Kubernetes MCP server in `deps/` is modified to use `loadFromCluster` for proper initialization in a Pod environment. > - The MCP servers in `deps/` are included to be packaged directly into the container image rather than downloaded at runtime. 4. Configure your Kubernetes access: - Ensure your kubeconfig is properly set up - The agent will use your current kubectl context 5. Start the application: ```bash chainlit run src/main.py ``` ### Helm Chart Deployment 1. Build and push the Docker image: ```bash # Build the Docker image docker build -t /kubernetes-ai-ops-agent: . # Push the image to your container registry docker push /kubernetes-ai-ops-agent: ``` 2. Create a `customized.values.yaml` file with your specific configuration: ```yaml # Customized values for kubernetes-ai-ops-agent image: repository: /kubernetes-ai-ops-agent # Configure the secrets secrets: data: # Option 1: For Azure OpenAI configuration AZURE_OPENAI_ENDPOINT: "https://.openai.azure.com/" AZURE_OPENAI_API_KEY: "" AZURE_OPENAI_MODEL: "" OPENAI_API_VERSION: "" # Option 2: For standard OpenAI configuration # OPENAI_API_KEY: "" # OPENAI_MODEL: "" # e.g., "gpt-4o" # Prometheus configuration PROMETHEUS_URL: "http://.:9090" ``` 3. Install the chart using your customized values: ```bash cd deploy/helm helm install kubernetes-ai-ops ./kubernetes-ai-ops-agent -f customized.values.yaml ``` 4. Access the application using port-forward: ```bash # Port-forward the service to access it locally # Use the namespace where you installed the chart (default used here as example) kubectl port-forward svc/kubernetes-ai-ops-agent 9000:9000 -n # Now you can access the web interface at http://localhost:9000 ``` > **Note**: Replace `` with the namespace where you installed the Helm chart. > If you installed without specifying a namespace (`-n` flag), the current context's > namespace will be used (usually "default"). ## Architecture The application is built with the following components: - **Chainlit Interface**: Web UI for interacting with the AI assistant - **Agent Layer**: Processes natural language, plans operations, and executes them - **MCP Servers**: - **Kubernetes MCP Server**: Handles Kubernetes API operations - **Prometheus MCP Server**: Provides monitoring metrics and data ## Model Support This project currently supports only OpenAI models. You can configure the agent to use either: - OpenAI API directly - Azure OpenAI API Configuration details for both options are provided in the deployment settings. ## Usage Examples ### Basic Cluster Information ``` User: "Show me all pods in the default namespace" ``` ### Troubleshooting ``` User: "Why is my pod in CrashLoopBackOff state?" ``` ## Project Structure - `src/`: Main application code - `main.py`: Main entry point for the Chainlit application - `chainlit_session_manager.py`: Manages Chainlit user sessions - `chainlit_session_storage.py`: Handles session data storage - `interfaces.py`: Defines interfaces and abstractions - `kubernetes_ai_ops_agent_provider.py`: Provider implementation for Kubernetes operations - `mcp_server_provider_impl.py`: Implementation for MCP server provider - `openai_client_factory_impl.py`: Factory for OpenAI client configuration - `deps/`: Dependencies and

Kubernetes AI Ops Agent

Tools