Environment Variables

All environment variables injected into agent pods by the ark-operator — operator-injected and user-configurable.

All environment variables available to agent pods.

Operator-injected variables

These are set by the operator based on the ArkAgent and ArkTeam spec. Do not set these manually — the operator manages them.

Variable	Description
`AGENT_MODEL`	Model ID from `spec.model` (e.g. `llama3.2`, `gpt-4o`)
`AGENT_SYSTEM_PROMPT`	Resolved system prompt (inline or from ConfigMap, merged with `ArkSettings` fragments)
`AGENT_MCP_SERVERS`	JSON array of `{name, url, headers}` — one entry per MCP server in `spec.mcpServers`
`AGENT_WEBHOOK_TOOLS`	JSON array of inline webhook tool definitions from `spec.tools`
`AGENT_MAX_TOKENS`	From `spec.limits.maxTokensPerCall` (default: `8000`)
`AGENT_TIMEOUT_SECONDS`	From `spec.limits.timeoutSeconds` (default: `120`)
`AGENT_MAX_CONCURRENT_TASKS`	From `spec.limits.maxConcurrentTasks` (default: `5`)
`TASK_QUEUE_URL`	Redis connection string for the task queue (e.g. `redis://redis.ark-system.svc.cluster.local:6379`)
`STREAM_CHANNEL_URL`	Redis connection string for the streaming channel. Defaults to `TASK_QUEUE_URL`.
`AGENT_TEAM_ROUTES`	(ArkTeam context) JSON map of `role → queueURL` for `delegate()` tool routing

Memory variables (operator-injected when `memoryRef` is set)

Variable	Description
`AGENT_MEMORY_BACKEND`	`in-context`, `redis`, or `vector-store`
`AGENT_MEMORY_REDIS_URL`	Redis connection URL (from the referenced Secret, redis backend)
`AGENT_MEMORY_REDIS_TTL`	TTL in seconds
`AGENT_MEMORY_REDIS_MAX_ENTRIES`	Entry cap
`AGENT_MEMORY_VECTOR_STORE_PROVIDER`	`qdrant`, `pinecone`, or `weaviate`
`AGENT_MEMORY_VECTOR_STORE_ENDPOINT`	Base URL
`AGENT_MEMORY_VECTOR_STORE_COLLECTION`	Collection name
`AGENT_MEMORY_VECTOR_STORE_API_KEY`	From the referenced Secret (if set)
`AGENT_MEMORY_VECTOR_STORE_TTL`	TTL in seconds

User-configurable variables

Set these via the apiKeys.existingSecret Kubernetes Secret, or via agentExtraEnv in Helm values. They are forwarded to every agent pod.

LLM provider credentials

Variable	Description
`ANTHROPIC_API_KEY`	API key for the Anthropic provider. Required when any agent uses a `claude-*` model.
`OPENAI_API_KEY`	API key for the OpenAI provider. Required when any agent uses a `gpt-/o` model. Also set to any non-empty value (e.g. `ollama`) when using Ollama.
`OPENAI_BASE_URL`	Override the OpenAI-compatible API endpoint. Set to `http://ollama.svc:11434/v1` for Ollama, or any other compatible endpoint.

Provider selection

Variable	Description
`AGENT_PROVIDER`	Override provider selection: `auto` (default), `anthropic`, `openai`, or `mock`. Use `auto` to let the agent runtime infer from the model name.

Observability

Variable	Description
`OTEL_EXPORTER_OTLP_ENDPOINT`	OTel collector endpoint (e.g. `http://jaeger.monitoring.svc:4318`). Enables distributed tracing and metrics export when set.
`OTEL_SERVICE_NAME`	Service name reported in traces. Default: `ark-agent`.

Setting user-configurable variables

Via Helm `agentExtraEnv` (recommended — cluster-wide)

helm upgrade ark-operator arkonis/ark-operator \
  --set agentExtraEnv[0].name=AGENT_PROVIDER,agentExtraEnv[0].value=openai \
  --set agentExtraEnv[1].name=OPENAI_BASE_URL,agentExtraEnv[1].value=http://ollama.svc:11434/v1 \
  --set agentExtraEnv[2].name=OPENAI_API_KEY,agentExtraEnv[2].value=ollama

Via Kubernetes Secret

kubectl create secret generic ark-api-keys \
  --from-literal=OPENAI_API_KEY=sk-... \
  --from-literal=ANTHROPIC_API_KEY=sk-ant-... \
  --namespace my-org

Reference the secret in Helm:

helm upgrade ark-operator arkonis/ark-operator \
  --set apiKeys.existingSecret=ark-api-keys