Struct CompletionModel 

pub struct CompletionModel {
    pub model: String,
    pub prompt_caching: bool,
    pub guardrail: Option<GuardrailConfiguration>,
    /* private fields */
}

Fields

model: String

prompt_caching: bool

When enabled, cache checkpoints are inserted into Converse API requests to take advantage of Bedrock prompt caching. A checkpoint is placed after the system prompt and after the last message in the chat history. Disabled by default.

guardrail: Option<GuardrailConfiguration>

Guardrail applied to every Converse request from this model, if any. Set through CompletionModel::with_guardrail.

Implementations

impl CompletionModel

pub fn new(client: Client, model: impl Into<String>) -> Self

pub fn with_prompt_caching(self) -> Self

Enable Bedrock prompt caching for this model.

When enabled, CachePoint blocks are inserted after the serialized system content and after the final messages entry in each Converse API request. This allows Bedrock to cache and reuse repeated prompt prefixes, reducing both latency and input token costs.

This currently covers the system and messages request fields only. Some Bedrock models also support caching tools, but that is not wired up here yet.

Cacheability and token thresholds are model-specific. If the cached prefix is too short or the model does not support caching for a given field, Bedrock ignores the checkpoint. See the Bedrock prompt caching support table for current limits and field support.

pub fn with_guardrail( self, identifier: impl Into<String>, version: impl Into<String>, trace: GuardrailTrace, ) -> Self

Apply a Bedrock guardrail to every Converse request this model issues.

identifier is the guardrail id or ARN and version its version (or DRAFT). When trace is enabled, Bedrock returns its assessment on InternalConverseOutput::trace — the only place it explains why a turn came back with StopReason::GuardrailIntervened, which the normalized response reports as a content filter and nothing more. Reach it through raw_completion.

impl CompletionModel

pub async fn raw_completion( &self, completion_request: CompletionRequest, ) -> Result<AwsConverseOutput, CompletionError>

Execute a completion and return Bedrock’s own Converse output.

This is the escape hatch for fields rig does not normalize; completion::CompletionModel::completion calls it and maps the result, so there is exactly one request either way.

impl CompletionModel

pub async fn raw_stream( &self, completion_request: CompletionRequest, ) -> Result<RawStreamingResult<BedrockStreamingResponse>, CompletionError>

Open a stream whose terminal record stays Bedrock’s own response type.

Trait Implementations

impl Clone for CompletionModel

fn clone(&self) -> CompletionModel

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl CompletionModel for CompletionModel

async fn completion( &self, completion_request: CompletionRequest, ) -> Result<CompletionResponse, CompletionError>

Generates a completion response for the given completion request.

async fn stream( &self, request: CompletionRequest, ) -> Result<StreamingCompletionResponse, CompletionError>

Streams a completion response for the given completion request.

fn completion_request( &self, prompt: impl Into<Message>, ) -> CompletionRequestBuilder<Self>

where Self: Sized + Clone,

Generates a completion request builder for the given prompt.

fn capabilities(&self) -> ProviderCapabilities

Provider behavior a runtime should account for when preparing requests.