Quickstart
Deploy the TalkOps Kubernetes Agent (k8s-autopilot), connect the web UI, configure your LLM provider credentials, and run your first live cluster task in under five minutes.
Prerequisites
- Docker & Docker Compose: Docker Engine 24+ and Docker Compose v2+
- Kubernetes access: A valid
~/.kube/configpointing to a local (kind,minikube,k3d, Docker Desktop) or remote cluster - LLM API key: An API key from at least one supported provider (Google Gemini, OpenAI, Anthropic, OpenRouter, AWS Bedrock, or Ollama)
Install and run your first task
Execute the installer script in your terminal:
curl -LsSf https://raw.githubusercontent.com/talkops-ai/k8s-autopilot/main/scripts/install.sh | bash
The script verifies Docker, Docker Compose, port availability (8888 for Web UI, 10102 for the A2A agent server), and ~/.kube/config, then pulls the images and starts the stack in ~/.k8s-autopilot/.
Run the installer inside WSL 2 (Windows Subsystem for Linux) so Docker containers can access your kubeconfig and cluster network.
Navigate to http://localhost:8888 in your browser. On the Connect to Agents screen, click the Quick Connect preset for K8s Autopilot (http://localhost:10102):
Once connected, the agent card displays its discovered A2A capabilities. Click Setup to configure your model provider credentials:
In Settings → Auth & Keys, select a provider row to add your API key:
| Provider | Environment key | Where to generate |
|---|---|---|
| Google Gemini | GEMINI_API_KEY | Google AI Studio |
| OpenAI | OPENAI_API_KEY | OpenAI Platform |
| Anthropic | ANTHROPIC_API_KEY | Anthropic Console |
Configured keys take effect immediately—you can switch across 20+ providers and models from the UI at any time without restarting containers. See Model Providers for the full list.
Click ← Back to Chat in the top right corner, then click Launch → on the agent card. Enter your first prompt:
List all pods in the default namespace and tell me if any are unhealthy
The Supervisor routes your request to the K8s Operator, which queries your cluster via kubernetes-mcp-server and streams back pod health. For any operation that modifies cluster state, k8s-autopilot renders an interactive A2UI approval card and waits for your confirmation.
Settings UI overview
All runtime settings are managed directly from the six panels in the Settings sidebar:
| Panel | What it configures |
|---|---|
| Auth & Keys | API keys, custom base URLs, and default models for 20+ LLM providers plus Tavily web search |
| Skills | Operational SKILL.md playbooks loaded by each operator (helm-operation, argocd-gitops, etc.) |
| Plugins | Remote plugin marketplaces and installed extension bundles (skills, subagents, MCP servers) |
| MCP Servers | Built-in and custom MCP server endpoints, health probes, and per-server enable/disable toggles |
| Traces | Real-time LangSmith tracing (LANGSMITH_API_KEY, project name, endpoint) |
| Runtime Config | Checkpointer storage hot-swap (SQLite vs. PostgreSQL), kubeconfig context, and observability URLs |
View Traces and Runtime Config screenshots
Alternative installation methods
Manual Docker Compose
If you prefer to manage the compose stack directly without the installer script:
mkdir -p ~/.k8s-autopilot && cd ~/.k8s-autopilot
curl -fsSL https://raw.githubusercontent.com/talkops-ai/k8s-autopilot/main/docker-compose.all-in-one.yml \
-o docker-compose.yml
curl -fsSL https://raw.githubusercontent.com/talkops-ai/k8s-autopilot/main/.env.example \
-o .env
docker compose up -d
Run from source (development)
Requires Python 3.12+, uv, and Node.js 20+:
git clone https://github.com/talkops-ai/k8s-autopilot.git
cd k8s-autopilot
# 1. Start the backend A2A agent server on port 10102
uv venv --python 3.12 && source .venv/bin/activate
uv pip install -e .
k8s-autopilot --host 0.0.0.0 --port 10102
In a second terminal, start the web UI on port 8888:
cd client-ui
npm install
npm run dev
Common lifecycle commands
Manage the installed Docker Compose stack from ~/.k8s-autopilot:
# View container status
docker compose -f ~/.k8s-autopilot/docker-compose.yml ps
# Tail agent logs
docker compose -f ~/.k8s-autopilot/docker-compose.yml logs -f k8s-autopilot
# Pull latest images and restart
docker compose -f ~/.k8s-autopilot/docker-compose.yml pull && \
docker compose -f ~/.k8s-autopilot/docker-compose.yml up -d
Example prompts to try
| Domain | Example prompt |
|---|---|
| Cluster health | "Check all namespaces for pods in CrashLoopBackOff or Pending state and explain the root cause" |
| Helm lifecycle | "Search Bitnami for Redis, validate values for a replica count of 2, and run a dry-run install in namespace cache" |
| GitOps (ArgoCD) | "List all ArgoCD applications, check if any are OutOfSync, and show the manifest diff" |
| Observability | "Query Prometheus for 5xx error rates by service over the last 30 minutes and correlate with Loki error logs" |
| Goal & rubric | "Deploy nginx via Helm in the demo namespace and verify all pods reach Ready with a rubric check" |
Next steps
Learn how Manual, Auto, and YOLO governance modes and the 5 security layers protect your cluster.
Configure SQLite/PostgreSQL persistence, cluster connectivity, and observability endpoints.
Deep dive into the Helm, App, K8s, and Observability operators and cross-operator handoffs.
Extend k8s-autopilot with community plugins, custom skills, and external MCP servers.