Bedrock Agents

A Bedrock Agent component provisions an AWS Bedrock agent as part of a StationOps infrastructure configuration. The editor separates the agent definition from actions, knowledge bases and guardrails, prompt overrides, access, and environment settings.

Agent Settings

1. Add Bedrock Agent to the environment components.
2. Enter a stable name and clear instructions.
3. Choose the foundation model, idle session TTL, orchestration mode, collaboration mode, and whether the agent prepares automatically.
4. Review the model selection and encryption key separately for every environment.

Actions

Action groups can use a Lambda service or return control to the caller. Define the action with an inline OpenAPI schema, an S3 schema reference, or a function schema. When Lambda is used, select a function-target service from the same configuration.

Keep action schemas narrow, validate every input in the executor, and design write operations so they can be authorised and audited independently of the model.

Knowledge Bases and Guardrails

Attach only the knowledge bases the agent needs and describe their purpose. If a guardrail is used, record both its identifier and version. Treat retrieval content as untrusted input and keep sensitive data out of sources that are broader than the agent's audience.

Prompt Overrides and Memory

Prompt overrides can target preprocessing, orchestration, post-processing, or knowledge-base response generation. Use the default prompts until a measured requirement justifies an override. If session-summary memory is enabled, choose a retention period that matches the data policy.

Connect Applications

Configure the agent ID and alias ID environment-variable names, then choose the services that should receive them. Use the Access tab for component permissions and verify the calling service has only the Bedrock operations it needs.