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

conversation/anthropic/anthropic.go

@@ -51,10 +51,16 @@ func (a *Anthropic) Init(ctx context.Context, meta conversation.Metadata) error
 	// Resolve model via central helper (uses metadata, then env var, then default)
 	model := conversation.GetAnthropicModel(m.Model)
 
-	llm, err := anthropic.New(
+	options := []anthropic.Option{
 		anthropic.WithModel(model),
 		anthropic.WithToken(m.Key),
-	)
+	}
+
+	if httpClient := conversation.BuildHTTPClient(m.HttpClientTimeout, m.IdleConnectionTimeout); httpClient != nil {
+		options = append(options, anthropic.WithHTTPClient(httpClient))
+	}
+
+	llm, err := anthropic.New(options...)
 	if err != nil {
 		return err
 	}

conversation/anthropic/metadata.yaml

@@ -34,3 +34,17 @@ metadata:
       A time-to-live value for a prompt cache to expire. Uses Golang durations
     type: string
     example: '10m'
+  - name: httpClientTimeout
+    required: false
+    description: |
+      Maximum duration for HTTP client requests to the Anthropic API.
+    type: duration
+    example: '30s'
+    default: '0s'
+  - name: idleConnectionTimeout
+    required: false
+    description: |
+      Maximum duration an idle HTTP connection can remain open before being closed.
+    type: duration
+    example: '30s'
+    default: '90s'

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

@@ -19,7 +19,6 @@ import (
 	"io"
 
 	"github.com/tmc/langchaingo/llms"
-	"google.golang.org/protobuf/types/known/anypb"
 
 	"github.com/dapr/components-contrib/metadata"
 )
@@ -36,23 +35,22 @@ 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"`
+	Model                      *string           `json:"model"`
 }
 
 type Response struct {
-	ConversationContext string   `json:"conversationContext"`
-	Outputs             []Result `json:"outputs"`
+	Outputs []Result `json:"outputs"`
+	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, md.HttpClientTimeout, md.IdleConnectionTimeout)
 	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/deepseek/metadata.yaml

@@ -41,3 +41,17 @@ metadata:
       Max tokens for each request
     type: number
     example: "2048"
+  - name: httpClientTimeout
+    required: false
+    description: |
+      Maximum duration for HTTP client requests to the Deepseek API.
+    type: duration
+    example: '30s'
+    default: '0s'
+  - name: idleConnectionTimeout
+    required: false
+    description: |
+      Maximum duration an idle HTTP connection can remain open before being closed.
+    type: duration
+    example: '30s'
+    default: '90s'

conversation/echo/echo.go

@@ -51,7 +51,9 @@ func (e *Echo) Init(ctx context.Context, meta conversation.Metadata) error {
 		return err
 	}
 
-	e.model = r.Model
+	if r.Model != nil {
+		e.model = *r.Model
+	}
 
 	return nil
 }
@@ -66,8 +68,7 @@ func (e *Echo) GetComponentMetadata() (metadataInfo metadata.MetadataMap) {
 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
 	}
 
@@ -139,6 +140,8 @@ func (e *Echo) Converse(ctx context.Context, r *conversation.Request) (res *conv
 		}
 	}
 
+	responseContent := strings.Join(contentFromMessaged, "\n")
+
 	stopReason := "stop"
 	if len(toolCalls) > 0 {
 		stopReason = "tool_calls"
@@ -148,7 +151,7 @@ func (e *Echo) Converse(ctx context.Context, r *conversation.Request) (res *conv
 		FinishReason: stopReason,
 		Index:        0,
 		Message: conversation.Message{
-			Content: strings.Join(contentFromMessaged, "\n"),
+			Content: responseContent,
 		},
 	}
 
@@ -161,9 +164,17 @@ func (e *Echo) Converse(ctx context.Context, r *conversation.Request) (res *conv
 		Choices:    []conversation.Choice{choice},
 	}
 
+	// allow per request model overrides
+	var modelName string
+	if r.Model != nil && *r.Model != "" {
+		modelName = *r.Model
+	} else {
+		modelName = e.model
+	}
+
 	res = &conversation.Response{
-		ConversationContext: r.ConversationContext,
-		Outputs:             []conversation.Result{output},
+		Outputs: []conversation.Result{output},
+		Model:   modelName,
 	}
 
 	return res, nil

conversation/googleai/googleai.go

@@ -50,16 +50,13 @@ 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/"),
-	}
-	llm, err := openai.New(
-		opts...,
-	)
+	// endpoint from https://ai.google.dev/gemini-api/docs/openai
+	const endpoint = "https://generativelanguage.googleapis.com/v1beta/openai/"
+	opts := conversation.BuildOpenAIClientOptions(model, key, endpoint, md.HttpClientTimeout, md.IdleConnectionTimeout)
+
+	llm, err := openai.New(opts...)
 	if err != nil {
 		return err
 	}

conversation/googleai/metadata.yaml

@@ -34,3 +34,17 @@ metadata:
       A time-to-live value for a prompt cache to expire. Uses Golang durations
     type: string
     example: '10m'
+  - name: httpClientTimeout
+    required: false
+    description: |
+      Maximum duration for HTTP client requests to the Google AI API.
+    type: duration
+    example: '30s'
+    default: '0s'
+  - name: idleConnectionTimeout
+    required: false
+    description: |
+      Maximum duration an idle HTTP connection can remain open before being closed.
+    type: duration
+    example: '30s'
+    default: '90s'

conversation/huggingface/huggingface.go

@@ -54,19 +54,14 @@ 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, m.HttpClientTimeout, m.IdleConnectionTimeout)
 
 	llm, err := openai.New(options...)
 	if err != nil {

conversation/huggingface/metadata.yaml

@@ -40,3 +40,17 @@ metadata:
       A time-to-live value for a prompt cache to expire. Uses Golang durations
     type: string
     example: '10m'
+  - name: httpClientTimeout
+    required: false
+    description: |
+      Maximum duration for HTTP client requests to the HuggingFace API.
+    type: duration
+    example: '30s'
+    default: '0s'
+  - name: idleConnectionTimeout
+    required: false
+    description: |
+      Maximum duration an idle HTTP connection can remain open before being closed.
+    type: duration
+    example: '30s'
+    default: '90s'