AWS describes deploying a music workflow with three agents using the Amazon Bedrock AgentCore Runtime Instances service
AWS's guide demonstrates a music workflow with three agents: one generates audio on a GPU, another edits it, and a third reviews it. They share files on a single instance; according to the company, the Amazon Bedrock AgentCore Runtime Instances service supports sessions of up to 14 days.
AWS has published a guide for deploying a music workflow with three specialized agents using the Amazon Bedrock AgentCore Runtime Instances service. The example uses Python and the Strands Agents framework. It includes creating a capacity provider, deploying individual agents, and connecting them via a shared session. The result is meant to be a playable .wav file and three messages explaining the agents' decisions.
The composition agent uses the Claude Sonnet 4.6 model to prepare the music brief and uses the ACE-Step model to generate audio on the instance's GPU. The mastering agent reads the file from shared storage, measures its parameters, and uses the Claude Sonnet 4.6 model to propose audio adjustments. After applying them, it measures the result again. The review agent independently verifies the target parameters and compares the audio against the studio's catalog; if a match is found, it can ask the composition agent for an alternative.
According to AWS, the Amazon Bedrock AgentCore Runtime Instances service supports sessions of up to 14 days, persistent Amazon EBS storage, and GPUs on supported instance families. The MicroVM variant supports sessions of at most eight hours and does not provide a GPU. Agents with the same capacity provider and runtimeSessionId can share a single instance as well as files. The compute instances run in the customer's account; AWS Savings Plans and On-Demand Capacity Reservations can be used. For details, see the source article.
Why it matters
The example connects audio generation, editing, and independent measurement through an actual shared file. For multi-day work, it's essential that files and decision history persist so agents can build on previous outputs. Checking against the studio's catalog provides a concrete mechanism for requesting a new version of the recording.
Two audiences, two different impacts
What this means
For individuals
Developers get an example of how to connect local audio generation on a GPU with other agents that work with the same file and retain session history.
For a business
The team can split music production across independently deployable agents and update one without affecting the others. Operation uses compute instances in the company's account, which must be factored into the cost of a multi-day workflow.
ProcessesCheck the original
Event sources
only one source so far · 1 publisher, 0 independent. We count feeds from the same owner only once.