Skip to content
View as markdown Ask ChatGPT Ask Claude

AIAgentModelConfiguration#

Description#

Resolves the configuration parameters to be applied for a given Agent and Model

Fields#

The AIAgentModelConfiguration table has the following fields:

Field Description Key Mandatory Comments
Agent The Agent that the resolved configuration applies to Yes No
Model The Model that the resolved configuration applies to Yes Yes
Configuration The configuration that establishes the parameters to be applied No Yes

Table: AIAgentModelConfiguration fields

Precedence Keys#

The search precedence rules are the following:

  1. Agent, Model
  2. Model

An agent-specific configuration overrides the default configuration for the same model.

Amazon Bedrock global inference profiles

The Model key is matched against the model identifier configured on the AI Agent. A global inference profile routes requests across supported AWS Regions, but its ARN still includes the source Region (for example, eu-north-1). A profile ARN from another source Region is a different model identifier and does not use this row's defaults. Create a model-level row for each identifier that needs the same defaults. For more information, see Amazon Bedrock global cross-Region inference ↗.

Configure Model Parameters#

  1. Go to Administration > Tables > Smart Tables.
  2. Open the AIAgentModelConfiguration SmartTable.
  3. Create or edit a row.
  4. Select an Agent when the configuration applies only to that agent. Leave it empty to configure a model-level default.
  5. Enter the Model identifier used by the agent.
  6. Enter the required Configuration JSON object and save the row.

During AI Agent synchronization, the system resolves the matching configuration and applies its parameters to the agent in the runtime.

Standard Configuration#

The standard installation seeds the following configuration for Amazon Bedrock global inference profiles:

  • The agent-specific row for CM Dashboard Builder Agent and arn:aws:bedrock:eu-north-1:211125333158:inference-profile/global.anthropic.claude-sonnet-5 uses {"thinking":false,"temperature":0.2,"maxTokens":128000}.
  • The model-level default for arn:aws:bedrock:eu-north-1:211125333158:inference-profile/global.anthropic.claude-sonnet-5 uses {"thinking":false,"temperature":0,"maxTokens":128000}.
  • The model-level default for arn:aws:bedrock:eu-north-1:211125333158:inference-profile/global.anthropic.claude-sonnet-5-5 uses {"thinking":false,"maxTokens":128000}.

Supported parameter names, values, and formats depend on the provider, model, and model version. Validate configuration against the provider's documentation before using different values. Unsupported parameters or invalid values can prevent the runtime from accepting the agent configuration.