Scaling

Manual scaling

In v1alpha1, replica count is set directly in spec.replicas:

spec:
  replicas: 5

The operator reconciles the backing pod count to match. Scale up or down by editing the ArkonisDeployment and applying it:

kubectl patch aodep research-agent --type=merge -p '{"spec":{"replicas":10}}'

Queue-depth autoscaling (planned)

CPU and memory are poor proxies for agent load. What matters is task queue depth — how many tasks are waiting to be processed.

Queue-depth-based autoscaling via KEDA is planned for v1beta1. The AgentScaler resource will allow defining scale-up/scale-down triggers based on Redis Streams queue length, with configurable thresholds and cooldown periods.

Resource limits

The spec.limits block controls per-agent resource consumption, not Kubernetes CPU/memory limits (those are set on the backing pods separately):

spec:
  limits:
    maxTokensPerCall: 8000
    maxConcurrentTasks: 5
    timeoutSeconds: 120

Field	Type	Default	Description
maxTokensPerCall	int	8000	Maximum tokens (input + output) per LLM API call.
maxConcurrentTasks	int	5	Maximum tasks a single agent pod will process simultaneously.
timeoutSeconds	int	120	Per-task timeout. The agent pod abandons the task and returns an error after this duration.

These values are injected as environment variables (AGENT_MAX_TOKENS, AGENT_TIMEOUT_SECONDS) into agent pods and enforced by the agent runtime.