Amazon Bedrock Agent
Does ZopNight manage Amazon Bedrock Agent?
Bedrock agents carry no meter of their own; every run bills through what the agent invokes: foundation-model tokens for reasoning, knowledge-base queries for retrieval, and Lambda charges in action groups. ZopNight discovers agents via its Bedrock provider on the 6-hour cycle and tracks the invocation cost each agent drives.
Rules that fire on Amazon Bedrock Agent
At a glance
| Field | Value |
|---|---|
| Scheduling notes | discovery and cost tracking only. |
A Bedrock agent orchestrates multi-step tasks using foundation models, action groups, and knowledge bases. Agents drive downstream model invocation and knowledge base query charges with every run.
An orchestrator that spends on your behalf
Creating and keeping an agent is free; running one is not, and the run cost is structurally multiplicative. A single agent invocation triggers a reasoning loop (the model plans, calls a tool, observes, plans again), and every iteration bills foundation-model tokens at the underlying model’s rates, with the agent’s instructions, tool schemas, and accumulated conversation riding along as input tokens each time. Knowledge-base lookups add retrieval charges, action groups add Lambda execution, and guardrails, if attached, meter each inspected exchange. One user question can quietly become a dozen model calls, which is why agent workloads surprise teams whose cost intuition was trained on single-shot completions.
Attribution for autonomous spenders
ZopNight’s Bedrock provider discovers agents on the 6-hour cycle and tracks the associated invocation cost from Cost Explorer or CUR 2.0 with usage trend analysis. Trend matters doubly for agents because their cost per run is not fixed: a prompt tweak, a new tool, or a model upgrade changes the loop’s depth and token appetite without any infrastructure change. A step change in an agent’s spend usually marks a behavioural regression: loops that run longer, retrieval that fires more often. Catching it in the cost series beats discovering it in the monthly invoice.
Loop-shaped overspend
The classic patterns: agents configured on premium models for tasks a lighter model handles, paying the flagship token rate on every iteration of every loop; runaway loops where a failing tool call sends the agent into retry-and-replan cycles; and oversized context, where verbose instructions and tool definitions are re-billed as input tokens on each reasoning step.
Agents in the Bedrock console
The Bedrock console’s Agents view lists agents with their model, action groups, and knowledge-base attachments. Reviewing each agent’s model choice and instruction length against its task, and watching per-agent invocation counts, keeps the orchestration layer’s autonomy inside a budget.