GuidedSellingModelCall
BETA: guided selling can still change in ways that break existing integrations. Backwards compatibility is not guaranteed while it is in beta.
One model request made for a turn, whether a customer waited for it or not.
type GuidedSellingModelCall {
id: ID!
createdAt: DateTime!
speculative: Boolean!
model: String!
effort: GuidedSellingEffort!
attempt: Int!
outcome: GuidedSellingModelCallOutcome!
inputTokens: Int
outputTokens: Int
cacheReadInputTokens: Int
cacheCreationInputTokens: Int
latencyMs: Int
}
Fields
GuidedSellingModelCall.id ● ID! non-null scalar
Identifier of the call. A call shows under every turn that was served the step it produced, so count each id once to total what a flow cost.
GuidedSellingModelCall.createdAt ● DateTime! non-null scalar
Moment the call finished.
GuidedSellingModelCall.speculative ● Boolean! non-null scalar
True for pre-generation and warming: nobody was waiting.
GuidedSellingModelCall.model ● String! non-null scalar
Identifier of the model that answered, as the vendor names it.
GuidedSellingModelCall.effort ● GuidedSellingEffort! non-null enum
Effort level the call ran at.
GuidedSellingModelCall.attempt ● Int! non-null scalar
1 for the first attempt, 2 for the retry after a rejected output.
GuidedSellingModelCall.outcome ● GuidedSellingModelCallOutcome! non-null enum
What the call led to.
GuidedSellingModelCall.inputTokens ● Int scalar
Input tokens billed at the full rate.
GuidedSellingModelCall.outputTokens ● Int scalar
Output tokens, thinking included.
GuidedSellingModelCall.cacheReadInputTokens ● Int scalar
Input tokens read from the vendor's prompt cache: the instruction and catalogue, at a fraction of the price.
GuidedSellingModelCall.cacheCreationInputTokens ● Int scalar
Input tokens written to the vendor's prompt cache: on the first call of a version, and again after about five minutes without calls, at a little more than the normal input rate.
GuidedSellingModelCall.latencyMs ● Int scalar
Duration of the vendor call in milliseconds.