Domain Agents›Kubernetes Agent›Configuration

Configuration

You can configure TalkOps Kubernetes Agent (k8s-autopilot) interactively through the Settings UI or declaratively via environment variables in ~/.k8s-autopilot/.env. Changes saved in the Settings UI take effect immediately without restarting containers.

Configuration precedence

When resolving any setting, k8s-autopilot evaluates sources in the following order (highest priority first):

Priority	Source	Behavior
1 (Highest)	Settings UI (Configuration database)	Applied immediately at runtime with zero container restarts
2	Environment variables (.env / shell)	Loaded at container startup as initial defaults
3 (Lowest)	Built-in framework defaults	Production-tested fallback values

Settings UI walkthrough

Open the Settings modal by clicking Setup on the agent card or the gear icon in the top navigation bar.

Auth & Keys

Manage API keys and custom base URLs for 20+ supported model providers. Each row displays the provider name, mapped environment variable, and active status ([configured] or [not set]).

Click any provider row to open its credential dialog:

• Save or rotate keys: Paste your API key and click Save Key. Keys are stored encrypted in the backend configuration store and masked in all API responses.
• Custom base URLs: Expand Advanced Settings to point the provider at an internal LiteLLM proxy, enterprise gateway, or VPC endpoint.

Plugins

Browse marketplaces, install community or private plugins, and manage enabled extensions:

• Discover: Browse plugins across all registered marketplaces and install with one click.
• Installed: Enable, disable, or uninstall installed plugins.
• Marketplaces: Register GitHub repositories (owner/repo), HTTPS marketplace.json catalogs, or local filesystem directories.
• Errors: Inspect diagnostic logs for any plugin that failed validation.

Traces

Configure LangSmith tracing to inspect every LangGraph node transition, prompt, tool call, and token count:

Setting	Environment variable	Default
Tracing enabled	LANGSMITH_TRACING	false
API key	LANGSMITH_API_KEY	—
Project name	LANGSMITH_PROJECT	k8s-autopilot
Endpoint URL	LANGSMITH_ENDPOINT	https://api.smith.langchain.com

Runtime Config

Configure state persistence, Kubernetes cluster contexts, GitOps endpoints, and observability URLs.

Backend storage

Choose where conversation threads, LangGraph checkpoints, and runtime settings are persisted:

Variable	Default	Description
CHECKPOINT_BACKEND	auto	Storage backend: sqlite, postgres, or auto (selects PostgreSQL when POSTGRES_URI is set)
SQLITE_FILE	k8s_autopilot.db	Local SQLite database file path (WAL mode enabled)
POSTGRES_URI	—	PostgreSQL connection URI (postgresql://user:pass@host:5432/dbname)

Kubernetes cluster connectivity

Configure the kubeconfig, context, default namespace, and sandbox execution provider:

Variable	Default	Description
KUBECONFIG	~/.kube/config	Path to the kubeconfig file mounted into the agent
KUBE_CONTEXT	(active context)	Kubernetes context name to target
KUBE_NAMESPACE	default	Default namespace when none is specified in a prompt
SANDBOX_PROVIDER	local	Execution environment for sandboxed commands (local, docker, or kubernetes)

Application delivery endpoints

Variable	Default	Description
ARGOCD_SERVER_URL	—	ArgoCD API server URL
ARGOCD_AUTH_TOKEN	—	ArgoCD bearer authentication token
ARGOCD_INSECURE	true	Skip TLS certificate verification for self-signed ArgoCD servers
TRAEFIK_API_URL	—	Traefik dashboard/API endpoint

Observability stack endpoints

Variable	Default	Description
PROMETHEUS_BASE_URL	http://prometheus-operated.monitoring.svc:9090	Prometheus query API URL
ALERTMANAGER_BASE_URL	http://alertmanager-operated.monitoring.svc:9093	Alertmanager API URL
LOKI_URL	http://localhost:3100	Grafana Loki HTTP API endpoint
TEMPO_BASE_URL	http://localhost:3200	Grafana Tempo HTTP API endpoint
OTEL_EXPORTER_OTLP_ENDPOINT	—	OpenTelemetry Collector OTLP receiver endpoint

Model & runtime environment variables

Variable	Default	Description
MODEL	gemini-3.7-flash	Active model identifier (e.g., claude-opus-4-7, gpt-4o, gemini-3.7-flash)
MODEL_PROVIDER	(auto-detected)	Explicit provider override (google_genai, anthropic, openai, ollama, etc.)
REASONING_EFFORT	medium	Extended thinking budget (low, medium, high, max)
MODEL_CONTEXT_LIMIT	(model default)	Override the context window token limit
APPROVAL_MODE	manual	Default governance mode (manual, auto, yolo)
RECURSION_LIMIT	50	Maximum LangGraph step recursion limit per turn
COMPACTION_TOKEN_BUDGET	100000	Token threshold that triggers automatic conversation compaction
HELM_WORKSPACE	./workspace/helm-charts	Local staging directory for generated Helm charts
GITHUB_PERSONAL_ACCESS_TOKEN	—	GitHub PAT used for committing generated charts and opening pull requests
TAVILY_API_KEY	—	Tavily API key for live web documentation search
LOG_LEVEL	INFO	Runtime log verbosity (DEBUG, INFO, WARNING, ERROR)
LOG_MODE	text	Log output format (text or json)

MCP transport and timeout variables

By default, built-in MCP servers run as local stdio subprocesses. To route specific MCP servers over HTTP in distributed Docker or Kubernetes deployments, set MCP_SERVERS:

MCP_SERVERS='[
  {"name": "helm_mcp_server", "url": "http://helm-svc:8080", "transport": "http"},
  {"name": "prometheus-mcp-server", "url": "http://prom-svc:8080", "transport": "http"}
]'

Variable	Default	Description
MCP_TIMEOUT	30	Per-tool execution timeout in seconds
MCP_TIMEOUT_TOTAL	600	Maximum total MCP operation timeout in seconds
MCP_TIMEOUT_CONNECT	300	MCP server initialization and connection timeout in seconds
MCP_DEFAULT_TRANSPORT	sse	Default transport for newly registered remote MCP servers (sse, stdio, http)

Automatic variable aliasing

To keep configuration DRY across multiple MCP servers, k8s-autopilot bidirectional-syncs common alias variables automatically:

• PROMETHEUS_BASE_URL ↔ PROMETHEUS_URL
• KUBECONFIG ↔ K8S_KUBECONFIG
• GOOGLE_API_KEY ↔ GEMINI_API_KEY
• LANGSMITH_API_KEY ↔ LANGCHAIN_API_KEY

Next steps

Model Providers

Configure 20+ LLM providers, reasoning budgets, Vertex AI, and local Ollama models.

MCP Tool Servers

Learn how .mcp.json files, JIT connections, and tool security tiers work.