diff --git a/chains/llm.go b/chains/llm.go
index e280d197d..9263ae066 100644
--- a/chains/llm.go
+++ b/chains/llm.go
@@ -19,6 +19,8 @@ type LLMChain struct {
 	Memory           schema.Memory
 	CallbacksHandler callbacks.Handler
 	OutputParser     schema.OutputParser[any]
+	// When enabled usesMultiplePrompts will not 'flatten' the prompt into a single message.
+	UseMultiPrompt bool
 
 	OutputKey string
 }
@@ -41,6 +43,7 @@ func NewLLMChain(llm llms.Model, prompt prompts.FormatPrompter, opts ...ChainCal
 		Memory:           memory.NewSimple(),
 		OutputKey:        _llmChainDefaultOutputKey,
 		CallbacksHandler: opt.CallbackHandler,
+		UseMultiPrompt:   false,
 	}
 
 	return chain
@@ -56,12 +59,17 @@ func (c LLMChain) Call(ctx context.Context, values map[string]any, options ...Ch
 		return nil, err
 	}
 
-	result, err := llms.GenerateFromMultiPrompt(ctx, c.LLM, chatMessagesToLLmMessageContent(promptValue.Messages()), getLLMCallOptions(options...)...)
+	var output string
+	if c.UseMultiPrompt {
+		output, err = llms.GenerateFromMultiPrompt(ctx, c.LLM, chatMessagesToLLmMessageContent(promptValue.Messages()), getLLMCallOptions(options...)...)
+	} else {
+		output, err = llms.GenerateFromSinglePrompt(ctx, c.LLM, promptValue.String(), getLLMCallOptions(options...)...)
+	}
 	if err != nil {
 		return nil, err
 	}
 
-	finalOutput, err := c.OutputParser.ParseWithPrompt(result, promptValue)
+	finalOutput, err := c.OutputParser.ParseWithPrompt(output, promptValue)
 	if err != nil {
 		return nil, err
 	}