feat: add resp format, timeout, model override to conversation api

conversation/anthropic/anthropic.go

@@ -61,8 +61,8 @@ func (a *Anthropic) Init(ctx context.Context, meta conversation.Metadata) error
 
 	a.LLM.Model = llm
 
-	if m.CacheTTL != "" {
-		cachedModel, cacheErr := conversation.CacheModel(ctx, m.CacheTTL, a.LLM.Model)
+	if m.ResponseCacheTTL != "" {
+		cachedModel, cacheErr := conversation.CacheResponses(ctx, m.ResponseCacheTTL, a.LLM.Model)
 		if cacheErr != nil {
 			return cacheErr
 		}

conversation/aws/bedrock/bedrock.go

@@ -42,8 +42,9 @@ type AWSBedrockMetadata struct {
 	AccessKey    string `json:"accessKey"`
 	SecretKey    string `json:"secretKey"`
 	SessionToken string `json:"sessionToken"`
-	Model        string `json:"model"`
-	CacheTTL     string `json:"cacheTTL"`
+
+	Model            string `json:"model"`
+	ResponseCacheTTL string `json:"responseCacheTTL" mapstructure:"responseCacheTTL" mapstructurealiases:"cacheTTL"`
 }
 
 func NewAWSBedrock(logger logger.Logger) conversation.Conversation {
@@ -83,8 +84,8 @@ func (b *AWSBedrock) Init(ctx context.Context, meta conversation.Metadata) error
 
 	b.LLM.Model = llm
 
-	if m.CacheTTL != "" {
-		cachedModel, cacheErr := conversation.CacheModel(ctx, m.CacheTTL, b.LLM.Model)
+	if m.ResponseCacheTTL != "" {
+		cachedModel, cacheErr := conversation.CacheResponses(ctx, m.ResponseCacheTTL, b.LLM.Model)
 		if cacheErr != nil {
 			return cacheErr
 		}

conversation/aws/bedrock/metadata.yaml

@@ -21,7 +21,10 @@ metadata:
   - name: model
     required: false
     description: |
-      The LLM to use. Defaults to Bedrock's default provider model from Amazon.
+      The model identifier or inference profile ARN to use. Defaults to Bedrock's default provider model from Amazon.
+      You can specify either:
+      - A model ID (e.g., "amazon.titan-text-express-v1") that supports on-demand throughput
+      - An inference profile ARN for models that require it (found in the AWS Bedrock console under "Cross-Region Inference")
     type: string
     example: 'amazon.titan-text-express-v1'
   - name: cacheTTL

conversation/converse.go

@@ -17,9 +17,9 @@ package conversation
 import (
 	"context"
 	"io"
+	"time"
 
 	"github.com/tmc/langchaingo/llms"
-	"google.golang.org/protobuf/types/known/anypb"
 
 	"github.com/dapr/components-contrib/metadata"
 )
@@ -36,23 +36,29 @@ type Conversation interface {
 
 type Request struct {
 	// Message can be user input prompt/instructions and/or tool call responses.
-	Message             *[]llms.MessageContent
-	Tools               *[]llms.Tool
-	ToolChoice          *string
-	Parameters          map[string]*anypb.Any `json:"parameters"`
-	ConversationContext string                `json:"conversationContext"`
-	Temperature         float64               `json:"temperature"`
-
-	// from metadata
-	Key       string   `json:"key"`
-	Model     string   `json:"model"`
-	Endpoints []string `json:"endpoints"`
-	Policy    string   `json:"loadBalancingPolicy"`
+	Message     *[]llms.MessageContent
+	Tools       *[]llms.Tool
+	ToolChoice  *string
+	Temperature float64 `json:"temperature"`
+
+	// Metadata fields that are separate from the actual component metadata fields
+	// that get passed to the LLM through the conversation.
+	// https://github.com/openai/openai-go/blob/main/chatcompletion.go#L3010
+	Metadata map[string]string `json:"metadata"`
+
+	ResponseFormatAsJSONSchema map[string]any `json:"responseFormatAsJsonSchema"`
+	PromptCacheRetention       time.Duration  `json:"promptCacheRetention"`
+	Model                      *string        `json:"model"`
+
+	// LlmTimeout specifies the max duration to wait for the LLM to complete a conversation request.
+	// Langchaingo timeout is respected, so if this is set, it will override the provider's timeout.
+	LlmTimeout time.Duration `json:"llmTimeout"`
 }
 
 type Response struct {
-	ConversationContext string   `json:"conversationContext"`
-	Outputs             []Result `json:"outputs"`
+	Outputs []Result `json:"outputs"`
+	Usage   *Usage   `json:"usage,omitempty"`
+	Model   string   `json:"model"`
 }
 
 type Result struct {

conversation/deepseek/deepseek.go

@@ -63,12 +63,7 @@ func (d *Deepseek) Init(ctx context.Context, meta conversation.Metadata) error {
 		md.Endpoint = defaultEndpoint
 	}
 
-	options := []openai.Option{
-		openai.WithModel(model),
-		openai.WithToken(md.Key),
-		openai.WithBaseURL(md.Endpoint),
-	}
-
+	options := conversation.BuildOpenAIClientOptions(model, md.Key, md.Endpoint)
 	llm, err := openai.New(options...)
 	if err != nil {
 		return err

conversation/deepseek/metadata.go

@@ -17,9 +17,10 @@ limitations under the License.
 
 package deepseek
 
+import "github.com/dapr/components-contrib/conversation"
+
 type DeepseekMetadata struct {
-	Key       string `json:"key"`
-	MaxTokens int    `json:"maxTokens"`
-	Model     string `json:"model"`
-	Endpoint  string `json:"endpoint"`
+	conversation.LangchainMetadata `json:",inline" mapstructure:",squash"`
+	Key                            string `json:"key"`
+	MaxTokens                      int    `json:"maxTokens"`
 }

conversation/echo/echo.go

@@ -51,7 +51,9 @@
 		return err
 	}
 
-	e.model = r.Model
+	if r.Model != nil {
+		e.model = *r.Model
+	}
 
 	return nil
 }
@@ -62,12 +64,23 @@
 	return
 }
 
+// approximateTokensFromWords estimates the number of tokens based on word count.
+// ref: https://help.openai.com/en/articles/4936856-what-are-tokens-and-how-to-count-them
+func approximateTokensFromWords(text string) int64 {
+	if text == "" {
+		return 0
+	}
+
+	// split on whitespace to count words
+	wordCount := len(strings.Fields(text))
+	return int64(wordCount)
+}
+
 // Converse returns one output per input message.
 func (e *Echo) Converse(ctx context.Context, r *conversation.Request) (res *conversation.Response, err error) {
 	if r == nil || r.Message == nil {
 		return &conversation.Response{
-			ConversationContext: r.ConversationContext,
-			Outputs:             []conversation.Result{},
+			Outputs: []conversation.Result{},
 		}, nil
 	}
 
@@ -121,24 +134,33 @@
 
 	// iterate over each message in the request to echo back the content in the response. We respond with the acummulated content of the message parts and tool responses
 	contentFromMessaged := make([]string, 0, len(*r.Message))
+	var promptTextLength int
 	for _, message := range *r.Message {
 		for _, part := range message.Parts {
 			switch p := part.(type) {
 			case llms.TextContent:
 				// append to slice that we'll join later with new line separators
 				contentFromMessaged = append(contentFromMessaged, p.Text)
+				promptTextLength += len(p.Text)
 			case llms.ToolCall:
 				// in case we added explicit tool calls on the request like on multi-turn conversations. We still append tool calls for each tool defined in the request.
 				toolCalls = append(toolCalls, p)
 			case llms.ToolCallResponse:
 				// show tool responses on the request like on multi-turn conversations
 				contentFromMessaged = append(contentFromMessaged, fmt.Sprintf("Tool Response for tool ID '%s' with name '%s': %s", p.ToolCallID, p.Name, p.Content))
+				promptTextLength += len(p.Content)
 			default:
 				return nil, fmt.Errorf("found invalid content type as input for %v", p)
 			}
 		}
 	}
 
+	responseContent := strings.Join(contentFromMessaged, "\n")
+
+	promptTokens := approximateTokensFromWords(promptTextLength)
+	completionTokens := approximateTokensFromWords(len(responseContent))
+	totalTokens := promptTokens + completionTokens
+
 	stopReason := "stop"
 	if len(toolCalls) > 0 {
 		stopReason = "tool_calls"
@@ -148,7 +170,7 @@
 		FinishReason: stopReason,
 		Index:        0,
 		Message: conversation.Message{
-			Content: strings.Join(contentFromMessaged, "\n"),
+			Content: responseContent,
 		},
 	}
 
@@ -161,9 +183,22 @@
 		Choices:    []conversation.Choice{choice},
 	}
 
+	// allow per request model overrides
+	modelName := e.model
+	if r.Model != nil && *r.Model != "" {
+		modelName = *r.Model
+	}
+
+	usage := &conversation.Usage{
+		CompletionTokens: completionTokens,
+		PromptTokens:     promptTokens,
+		TotalTokens:      totalTokens,
+	}
+
 	res = &conversation.Response{
-		ConversationContext: r.ConversationContext,
-		Outputs:             []conversation.Result{output},
+		Outputs: []conversation.Result{output},
+		Usage:   usage,
+		Model:   modelName,
 	}
 
 	return res, nil

conversation/googleai/googleai.go

@@ -50,13 +50,11 @@ func (g *GoogleAI) Init(ctx context.Context, meta conversation.Metadata) error {
 
 	// Resolve model via central helper (uses metadata, then env var, then default)
 	model := conversation.GetGoogleAIModel(md.Model)
+	key, _ := meta.GetProperty("key")
 
-	opts := []openai.Option{
-		openai.WithModel(model),
-		openai.WithToken(md.Key),
-		// endpoint from https://ai.google.dev/gemini-api/docs/openai
-		openai.WithBaseURL("https://generativelanguage.googleapis.com/v1beta/openai/"),
-	}
+	// endpoint from https://ai.google.dev/gemini-api/docs/openai
+	const endpoint = "https://generativelanguage.googleapis.com/v1beta/openai/"
+	opts := conversation.BuildOpenAIClientOptions(model, key, endpoint)
 	llm, err := openai.New(
 		opts...,
 	)
@@ -66,8 +64,8 @@ func (g *GoogleAI) Init(ctx context.Context, meta conversation.Metadata) error {
 
 	g.LLM.Model = llm
 
-	if md.CacheTTL != "" {
-		cachedModel, cacheErr := conversation.CacheModel(ctx, md.CacheTTL, g.LLM.Model)
+	if md.ResponseCacheTTL != "" {
+		cachedModel, cacheErr := conversation.CacheResponses(ctx, md.ResponseCacheTTL, g.LLM.Model)
 		if cacheErr != nil {
 			return cacheErr
 		}

conversation/huggingface/huggingface.go

@@ -54,29 +54,23 @@ func (h *Huggingface) Init(ctx context.Context, meta conversation.Metadata) erro
 
 	// Resolve model via central helper (uses metadata, then env var, then default)
 	model := conversation.GetHuggingFaceModel(m.Model)
-
 	endpoint := strings.Replace(defaultEndpoint, "{{model}}", model, 1)
 	if m.Endpoint != "" {
 		endpoint = m.Endpoint
 	}
 
 	// Create options for OpenAI client using HuggingFace's OpenAI-compatible API
 	// This is a workaround for issues with the native HuggingFace langchaingo implementation
-	options := []openai.Option{
-		openai.WithModel(model),
-		openai.WithToken(m.Key),
-		openai.WithBaseURL(endpoint),
-	}
-
+	options := conversation.BuildOpenAIClientOptions(model, m.Key, endpoint)
 	llm, err := openai.New(options...)
 	if err != nil {
 		return err
 	}
 
 	h.LLM.Model = llm
 
-	if m.CacheTTL != "" {
-		cachedModel, cacheErr := conversation.CacheModel(ctx, m.CacheTTL, h.LLM.Model)
+	if m.ResponseCacheTTL != "" {
+		cachedModel, cacheErr := conversation.CacheResponses(ctx, m.ResponseCacheTTL, h.LLM.Model)
 		if cacheErr != nil {
 			return cacheErr
 		}