diff --git a/pkg/cli/cmd.go b/pkg/cli/cmd.go
index 0ab667e..088b9f4 100644
--- a/pkg/cli/cmd.go
+++ b/pkg/cli/cmd.go
@@ -9,10 +9,16 @@ import (
 	"github.com/spf13/cobra"
 )
 
-// TODO: allow setting with flag
-const MAX_TOKENS = 256
+var (
+	maxTokens int
+)
 
 func init() {
+	inputCmds := []*cobra.Command{newCmd, promptCmd, replyCmd}
+	for _, cmd := range inputCmds {
+		cmd.Flags().IntVar(&maxTokens, "length", config.OpenAI.DefaultMaxLength, "Max response length in tokens")
+	}
+
 	rootCmd.AddCommand(
 		lsCmd,
 		newCmd,
@@ -268,7 +274,7 @@ var replyCmd = &cobra.Command{
 			response <- HandleDelayedResponse(receiver)
 		}()
 
-		err = CreateChatCompletionStream(config.OpenAI.DefaultModel, messages, MAX_TOKENS, receiver)
+		err = CreateChatCompletionStream(config.OpenAI.DefaultModel, messages, maxTokens, receiver)
 		if err != nil {
 			Fatal("%v\n", err)
 		}
@@ -348,7 +354,7 @@ var newCmd = &cobra.Command{
 			response <- HandleDelayedResponse(receiver)
 		}()
 
-		err = CreateChatCompletionStream(config.OpenAI.DefaultModel, messages, MAX_TOKENS, receiver)
+		err = CreateChatCompletionStream(config.OpenAI.DefaultModel, messages, maxTokens, receiver)
 		if err != nil {
 			Fatal("%v\n", err)
 		}
@@ -397,7 +403,7 @@ var promptCmd = &cobra.Command{
 
 		receiver := make(chan string)
 		go HandleDelayedResponse(receiver)
-		err := CreateChatCompletionStream(config.OpenAI.DefaultModel, messages, MAX_TOKENS, receiver)
+		err := CreateChatCompletionStream(config.OpenAI.DefaultModel, messages, maxTokens, receiver)
 		if err != nil {
 			Fatal("%v\n", err)
 		}
diff --git a/pkg/cli/config.go b/pkg/cli/config.go
index 41aa2a7..5e94dbb 100644
--- a/pkg/cli/config.go
+++ b/pkg/cli/config.go
@@ -12,6 +12,7 @@ type Config struct {
 	OpenAI struct {
 		APIKey string `yaml:"apiKey"`
 		DefaultModel string `yaml:"defaultModel"`
+		DefaultMaxLength int `yaml:"defaultMaxLength"`
 	} `yaml:"openai"`
 }