diff --git a/agents/conversational.go b/agents/conversational.go
index 5380d9072..fd12c4630 100644
--- a/agents/conversational.go
+++ b/agents/conversational.go
@@ -2,23 +2,19 @@ package agents
 
 import (
 	"context"
-	_ "embed"
 	"fmt"
 	"regexp"
 	"strings"
 
 	"github.com/tmc/langchaingo/callbacks"
 	"github.com/tmc/langchaingo/chains"
+	"github.com/tmc/langchaingo/i18n"
 	"github.com/tmc/langchaingo/llms"
 	"github.com/tmc/langchaingo/prompts"
 	"github.com/tmc/langchaingo/schema"
 	"github.com/tmc/langchaingo/tools"
 )
 
-const (
-	_conversationalFinalAnswerAction = "AI:"
-)
-
 // ConversationalAgent is a struct that represents an agent responsible for deciding
 // what to do or give the final output if the task is finished given a set of inputs
 // and previous steps taken.
@@ -34,6 +30,10 @@ type ConversationalAgent struct {
 	Tools []tools.Tool
 	// Output key is the key where the final output is placed.
 	OutputKey string
+	// FinalAnswer is the final answer in various languages.
+	FinalAnswer string
+	// Lang is the language the prompt will use.
+	Lang i18n.Lang
 	// CallbacksHandler is the handler for callbacks.
 	CallbacksHandler callbacks.Handler
 }
@@ -45,6 +45,7 @@ func NewConversationalAgent(llm llms.Model, tools []tools.Tool, opts ...Option)
 	for _, opt := range opts {
 		opt(&options)
 	}
+	options.loadConversationalTranslatable()
 
 	return &ConversationalAgent{
 		Chain: chains.NewLLMChain(
@@ -54,6 +55,8 @@ func NewConversationalAgent(llm llms.Model, tools []tools.Tool, opts ...Option)
 		),
 		Tools:            tools,
 		OutputKey:        options.outputKey,
+		FinalAnswer:      i18n.AgentsMustPhrase(options.lang, "conversational final answer"),
+		Lang:             options.lang,
 		CallbacksHandler: options.callbacksHandler,
 	}
 }
@@ -69,7 +72,7 @@ func (a *ConversationalAgent) Plan(
 		fullInputs[key] = value
 	}
 
-	fullInputs["agent_scratchpad"] = constructScratchPad(intermediateSteps)
+	fullInputs["agent_scratchpad"] = constructScratchPad(intermediateSteps, a.Lang)
 
 	var stream func(ctx context.Context, chunk []byte) error
 
@@ -84,7 +87,10 @@ func (a *ConversationalAgent) Plan(
 		ctx,
 		a.Chain,
 		fullInputs,
-		chains.WithStopWords([]string{"\nObservation:", "\n\tObservation:"}),
+		chains.WithStopWords([]string{
+			fmt.Sprintf("\n%s", i18n.AgentsMustPhrase(a.Lang, "observation")),
+			fmt.Sprintf("\n\t%s", i18n.AgentsMustPhrase(a.Lang, "observation")),
+		}),
 		chains.WithStreamingFunc(stream),
 	)
 	if err != nil {
@@ -117,22 +123,22 @@ func (a *ConversationalAgent) GetTools() []tools.Tool {
 	return a.Tools
 }
 
-func constructScratchPad(steps []schema.AgentStep) string {
+func constructScratchPad(steps []schema.AgentStep, lang i18n.Lang) string {
 	var scratchPad string
 	if len(steps) > 0 {
 		for _, step := range steps {
 			scratchPad += step.Action.Log
-			scratchPad += "\nObservation: " + step.Observation
+			scratchPad += fmt.Sprintf("\n%s %s", i18n.AgentsMustPhrase(lang, "observation"), step.Observation)
 		}
-		scratchPad += "\n" + "Thought:"
+		scratchPad += fmt.Sprintf("\n%s", i18n.AgentsMustPhrase(lang, "thought"))
 	}
 
 	return scratchPad
 }
 
 func (a *ConversationalAgent) parseOutput(output string) ([]schema.AgentAction, *schema.AgentFinish, error) {
-	if strings.Contains(output, _conversationalFinalAnswerAction) {
-		splits := strings.Split(output, _conversationalFinalAnswerAction)
+	if strings.Contains(output, a.FinalAnswer) {
+		splits := strings.Split(output, a.FinalAnswer)
 
 		finishAction := &schema.AgentFinish{
 			ReturnValues: map[string]any{
@@ -144,7 +150,9 @@ func (a *ConversationalAgent) parseOutput(output string) ([]schema.AgentAction,
 		return nil, finishAction, nil
 	}
 
-	r := regexp.MustCompile(`Action: (.*?)[\n]*Action Input: (.*)`)
+	action, actionInput := i18n.AgentsMustPhrase(a.Lang, "action"),
+		i18n.AgentsMustPhrase(a.Lang, "action input")
+	r := regexp.MustCompile(fmt.Sprintf(`%s (.*?)[\n]*%s (.*)`, action, actionInput))
 	matches := r.FindStringSubmatch(output)
 	if len(matches) == 0 {
 		return nil, nil, fmt.Errorf("%w: %s", ErrUnableToParseOutput, output)
@@ -155,15 +163,6 @@ func (a *ConversationalAgent) parseOutput(output string) ([]schema.AgentAction,
 	}, nil, nil
 }
 
-//go:embed prompts/conversational_prefix.txt
-var _defaultConversationalPrefix string //nolint:gochecknoglobals
-
-//go:embed prompts/conversational_format_instructions.txt
-var _defaultConversationalFormatInstructions string //nolint:gochecknoglobals
-
-//go:embed prompts/conversational_suffix.txt
-var _defaultConversationalSuffix string //nolint:gochecknoglobals
-
 func createConversationalPrompt(tools []tools.Tool, prefix, instructions, suffix string) prompts.PromptTemplate {
 	template := strings.Join([]string{prefix, instructions, suffix}, "\n\n")
 
diff --git a/agents/executor.go b/agents/executor.go
index 5835ad008..8f0e340c5 100644
--- a/agents/executor.go
+++ b/agents/executor.go
@@ -36,6 +36,7 @@ func NewExecutor(agent Agent, opts ...Option) *Executor {
 	for _, opt := range opts {
 		opt(&options)
 	}
+	options.loadExecutorTranslatable()
 
 	return &Executor{
 		Agent:                   agent,
diff --git a/agents/mrkl.go b/agents/mrkl.go
index 4bc577408..ec312be9a 100644
--- a/agents/mrkl.go
+++ b/agents/mrkl.go
@@ -9,16 +9,12 @@ import (
 
 	"github.com/tmc/langchaingo/callbacks"
 	"github.com/tmc/langchaingo/chains"
+	"github.com/tmc/langchaingo/i18n"
 	"github.com/tmc/langchaingo/llms"
 	"github.com/tmc/langchaingo/schema"
 	"github.com/tmc/langchaingo/tools"
 )
 
-const (
-	_finalAnswerAction = "Final Answer:"
-	_defaultOutputKey  = "output"
-)
-
 // OneShotZeroAgent is a struct that represents an agent responsible for deciding
 // what to do or give the final output if the task is finished given a set of inputs
 // and previous steps taken.
@@ -32,6 +28,10 @@ type OneShotZeroAgent struct {
 	Tools []tools.Tool
 	// Output key is the key where the final output is placed.
 	OutputKey string
+	// FinalAnswer is the final answer in various languages.
+	FinalAnswer string
+	// Lang is the language the prompt will use.
+	Lang i18n.Lang
 	// CallbacksHandler is the handler for callbacks.
 	CallbacksHandler callbacks.Handler
 }
@@ -46,6 +46,7 @@ func NewOneShotAgent(llm llms.Model, tools []tools.Tool, opts ...Option) *OneSho
 	for _, opt := range opts {
 		opt(&options)
 	}
+	options.loadMrklTranslatable()
 
 	return &OneShotZeroAgent{
 		Chain: chains.NewLLMChain(
@@ -55,6 +56,8 @@ func NewOneShotAgent(llm llms.Model, tools []tools.Tool, opts ...Option) *OneSho
 		),
 		Tools:            tools,
 		OutputKey:        options.outputKey,
+		FinalAnswer:      i18n.AgentsMustPhrase(options.lang, "mrkl final answer"),
+		Lang:             options.lang,
 		CallbacksHandler: options.callbacksHandler,
 	}
 }
@@ -70,8 +73,8 @@ func (a *OneShotZeroAgent) Plan(
 		fullInputs[key] = value
 	}
 
-	fullInputs["agent_scratchpad"] = constructMrklScratchPad(intermediateSteps)
-	fullInputs["today"] = time.Now().Format("January 02, 2006")
+	fullInputs["agent_scratchpad"] = constructMrklScratchPad(intermediateSteps, a.Lang)
+	fullInputs["today"] = time.Now().Format(i18n.AgentsMustPhrase(a.Lang, "today format"))
 
 	var stream func(ctx context.Context, chunk []byte) error
 
@@ -86,7 +89,10 @@ func (a *OneShotZeroAgent) Plan(
 		ctx,
 		a.Chain,
 		fullInputs,
-		chains.WithStopWords([]string{"\nObservation:", "\n\tObservation:"}),
+		chains.WithStopWords([]string{
+			fmt.Sprintf("\n%s", i18n.AgentsMustPhrase(a.Lang, "observation")),
+			fmt.Sprintf("\n\t%s", i18n.AgentsMustPhrase(a.Lang, "observation")),
+		}),
 		chains.WithStreamingFunc(stream),
 	)
 	if err != nil {
@@ -119,12 +125,12 @@ func (a *OneShotZeroAgent) GetTools() []tools.Tool {
 	return a.Tools
 }
 
-func constructMrklScratchPad(steps []schema.AgentStep) string {
+func constructMrklScratchPad(steps []schema.AgentStep, lang i18n.Lang) string {
 	var scratchPad string
 	if len(steps) > 0 {
 		for _, step := range steps {
 			scratchPad += "\n" + step.Action.Log
-			scratchPad += "\nObservation: " + step.Observation + "\n"
+			scratchPad += fmt.Sprintf("\n%s %s\n", i18n.AgentsMustPhrase(lang, "observation"), step.Observation)
 		}
 	}
 
@@ -132,8 +138,8 @@ func constructMrklScratchPad(steps []schema.AgentStep) string {
 }
 
 func (a *OneShotZeroAgent) parseOutput(output string) ([]schema.AgentAction, *schema.AgentFinish, error) {
-	if strings.Contains(output, _finalAnswerAction) {
-		splits := strings.Split(output, _finalAnswerAction)
+	if strings.Contains(output, a.FinalAnswer) {
+		splits := strings.Split(output, a.FinalAnswer)
 
 		return nil, &schema.AgentFinish{
 			ReturnValues: map[string]any{
@@ -143,7 +149,15 @@ func (a *OneShotZeroAgent) parseOutput(output string) ([]schema.AgentAction, *sc
 		}, nil
 	}
 
-	r := regexp.MustCompile(`Action:\s*(.+)\s*Action Input:\s(?s)*(.+)`)
+	action, actionInput, observation := i18n.AgentsMustPhrase(a.Lang, "action"),
+		i18n.AgentsMustPhrase(a.Lang, "action input"),
+		i18n.AgentsMustPhrase(a.Lang, "observation")
+	var r *regexp.Regexp
+	if strings.Contains(output, observation) {
+		r = regexp.MustCompile(fmt.Sprintf(`%s\s*(.+)\s*%s\s(?s)*(.+)%s`, action, actionInput, observation))
+	} else {
+		r = regexp.MustCompile(fmt.Sprintf(`%s\s*(.+)\s*%s\s(?s)*(.+)`, action, actionInput))
+	}
 	matches := r.FindStringSubmatch(output)
 	if len(matches) == 0 {
 		return nil, nil, fmt.Errorf("%w: %s", ErrUnableToParseOutput, output)
diff --git a/agents/mrkl_prompt.go b/agents/mrkl_prompt.go
index eb929febb..6fdedf424 100644
--- a/agents/mrkl_prompt.go
+++ b/agents/mrkl_prompt.go
@@ -8,29 +8,6 @@ import (
 	"github.com/tmc/langchaingo/tools"
 )
 
-const (
-	_defaultMrklPrefix = `Today is {{.today}}.
-Answer the following questions as best you can. You have access to the following tools:
-
-{{.tool_descriptions}}`
-
-	_defaultMrklFormatInstructions = `Use the following format:
-
-Question: the input question you must answer
-Thought: you should always think about what to do
-Action: the action to take, should be one of [ {{.tool_names}} ]
-Action Input: the input to the action
-Observation: the result of the action
-... (this Thought/Action/Action Input/Observation can repeat N times)
-Thought: I now know the final answer
-Final Answer: the final answer to the original input question`
-
-	_defaultMrklSuffix = `Begin!
-
-Question: {{.input}}
-{{.agent_scratchpad}}`
-)
-
 func createMRKLPrompt(tools []tools.Tool, prefix, instructions, suffix string) prompts.PromptTemplate {
 	template := strings.Join([]string{prefix, instructions, suffix}, "\n\n")
 
diff --git a/agents/markl_test.go b/agents/mrkl_test.go
similarity index 86%
rename from agents/markl_test.go
rename to agents/mrkl_test.go
index 3112f77ea..671a8b72e 100644
--- a/agents/markl_test.go
+++ b/agents/mrkl_test.go
@@ -4,6 +4,7 @@ import (
 	"testing"
 
 	"github.com/stretchr/testify/require"
+	"github.com/tmc/langchaingo/i18n"
 	"github.com/tmc/langchaingo/schema"
 )
 
@@ -38,7 +39,11 @@ func TestMRKLOutputParser(t *testing.T) {
 		},
 	}
 
-	a := OneShotZeroAgent{}
+	lang := i18n.EN
+	a := OneShotZeroAgent{
+		FinalAnswer: i18n.AgentsMustPhrase(lang, "mrkl final answer"),
+		Lang:        lang,
+	}
 	for _, tc := range testCases {
 		actions, finish, err := a.parseOutput(tc.input)
 		require.ErrorIs(t, tc.expectedErr, err)
diff --git a/agents/openai_functions_agent.go b/agents/openai_functions_agent.go
index 1ff621a29..533dbdf93 100644
--- a/agents/openai_functions_agent.go
+++ b/agents/openai_functions_agent.go
@@ -38,6 +38,7 @@ func NewOpenAIFunctionsAgent(llm llms.Model, tools []tools.Tool, opts ...Option)
 	for _, opt := range opts {
 		opt(&options)
 	}
+	options.loadOpenAIFunctionsTranslatable()
 
 	return &OpenAIFunctionsAgent{
 		LLM:              llm,
diff --git a/agents/options.go b/agents/options.go
index 946b48c4d..054b30382 100644
--- a/agents/options.go
+++ b/agents/options.go
@@ -2,12 +2,20 @@ package agents
 
 import (
 	"github.com/tmc/langchaingo/callbacks"
+	"github.com/tmc/langchaingo/i18n"
 	"github.com/tmc/langchaingo/memory"
 	"github.com/tmc/langchaingo/prompts"
 	"github.com/tmc/langchaingo/schema"
 	"github.com/tmc/langchaingo/tools"
 )
 
+type translatable struct {
+	promptPrefix       string
+	formatInstructions string
+	promptSuffix       string
+	outputKey          string
+}
+
 type Options struct {
 	prompt                  prompts.PromptTemplate
 	memory                  schema.Memory
@@ -15,10 +23,8 @@ type Options struct {
 	errorHandler            *ParserErrorHandler
 	maxIterations           int
 	returnIntermediateSteps bool
-	outputKey               string
-	promptPrefix            string
-	formatInstructions      string
-	promptSuffix            string
+	lang                    i18n.Lang
+	translatable
 
 	// openai
 	systemMessage string
@@ -29,40 +35,55 @@ type Options struct {
 // and executors.
 type Option func(*Options)
 
-func executorDefaultOptions() Options {
+func defaultOptions() Options {
 	return Options{
-		maxIterations: _defaultMaxIterations,
-		outputKey:     _defaultOutputKey,
-		memory:        memory.NewSimple(),
+		lang: i18n.DefaultLang,
 	}
 }
 
+func executorDefaultOptions() Options {
+	options := defaultOptions()
+	options.maxIterations = _defaultMaxIterations
+	options.memory = memory.NewSimple()
+	return options
+}
+
 func mrklDefaultOptions() Options {
-	return Options{
-		promptPrefix:       _defaultMrklPrefix,
-		formatInstructions: _defaultMrklFormatInstructions,
-		promptSuffix:       _defaultMrklSuffix,
-		outputKey:          _defaultOutputKey,
-	}
+	return defaultOptions()
 }
 
 func conversationalDefaultOptions() Options {
-	return Options{
-		promptPrefix:       _defaultConversationalPrefix,
-		formatInstructions: _defaultConversationalFormatInstructions,
-		promptSuffix:       _defaultConversationalSuffix,
-		outputKey:          _defaultOutputKey,
-	}
+	return defaultOptions()
 }
 
 func openAIFunctionsDefaultOptions() Options {
-	return Options{
-		systemMessage: "You are a helpful AI assistant.",
-		outputKey:     _defaultOutputKey,
-	}
+	return defaultOptions()
 }
 
-func (co Options) getMrklPrompt(tools []tools.Tool) prompts.PromptTemplate {
+func (co *Options) loadExecutorTranslatable() {
+	co.outputKey = i18n.AgentsMustPhrase(co.lang, "output key")
+}
+
+func (co *Options) loadMrklTranslatable() {
+	co.promptPrefix = i18n.AgentsMustLoad(co.lang, "mrkl_prompt_prefix.txt")
+	co.formatInstructions = i18n.AgentsMustLoad(co.lang, "mrkl_prompt_format_instructions.txt")
+	co.promptSuffix = i18n.AgentsMustLoad(co.lang, "mrkl_prompt_suffix.txt")
+	co.outputKey = i18n.AgentsMustPhrase(co.lang, "output key")
+}
+
+func (co *Options) loadConversationalTranslatable() {
+	co.promptPrefix = i18n.AgentsMustLoad(co.lang, "conversational_prompt_prefix.txt")
+	co.formatInstructions = i18n.AgentsMustLoad(co.lang, "conversational_prompt_format_instructions.txt")
+	co.promptSuffix = i18n.AgentsMustLoad(co.lang, "conversational_prompt_suffix.txt")
+	co.outputKey = i18n.AgentsMustPhrase(co.lang, "output key")
+}
+
+func (co *Options) loadOpenAIFunctionsTranslatable() {
+	co.systemMessage = i18n.AgentsMustPhrase(co.lang, "system message")
+	co.outputKey = i18n.AgentsMustPhrase(co.lang, "output key")
+}
+
+func (co *Options) getMrklPrompt(tools []tools.Tool) prompts.PromptTemplate {
 	if co.prompt.Template != "" {
 		return co.prompt
 	}
@@ -75,7 +96,7 @@ func (co Options) getMrklPrompt(tools []tools.Tool) prompts.PromptTemplate {
 	)
 }
 
-func (co Options) getConversationalPrompt(tools []tools.Tool) prompts.PromptTemplate {
+func (co *Options) getConversationalPrompt(tools []tools.Tool) prompts.PromptTemplate {
 	if co.prompt.Template != "" {
 		return co.prompt
 	}
@@ -161,6 +182,13 @@ func WithParserErrorHandler(errorHandler *ParserErrorHandler) Option {
 	}
 }
 
+// WithLang is an option for setting language the prompt will use.
+func WithLang(lang i18n.Lang) Option {
+	return func(co *Options) {
+		co.lang = lang
+	}
+}
+
 type OpenAIOption struct{}
 
 func NewOpenAIOption() OpenAIOption {
diff --git a/i18n/doc.go b/i18n/doc.go
new file mode 100644
index 000000000..6762de5b4
--- /dev/null
+++ b/i18n/doc.go
@@ -0,0 +1,6 @@
+// Package i18n provides internationalization for other packages,
+// currently the agents package.
+//   - Support for package agents: If non-English questions are inputted,
+//     the agents won't work well with the English prompt, so the
+//     corresponding-language prompt will be necessary.
+package i18n
diff --git a/i18n/lang.go b/i18n/lang.go
new file mode 100644
index 000000000..eaaef4269
--- /dev/null
+++ b/i18n/lang.go
@@ -0,0 +1,14 @@
+package i18n
+
+// Lang is the language type.
+type Lang int
+
+const (
+	// EN stands for English.
+	EN Lang = iota
+	// ZH stands for Simplified-Chinese.
+	ZH
+
+	// DefaultLang is the default language.
+	DefaultLang = EN
+)
diff --git a/i18n/load.go b/i18n/load.go
new file mode 100644
index 000000000..97cf3cac4
--- /dev/null
+++ b/i18n/load.go
@@ -0,0 +1,28 @@
+package i18n
+
+import (
+	"embed"
+	"fmt"
+	"io/fs"
+	"log"
+)
+
+//go:embed templates/*
+var tpls embed.FS
+
+func mustLoad(lang Lang, kindFolder, filename string) string {
+	langFolderMap := map[Lang]string{
+		EN: "en",
+		ZH: "zh",
+	}
+	langFolder, ok := langFolderMap[lang]
+	if !ok {
+		log.Panic("unknown language: ", lang)
+	}
+	filepath := fmt.Sprintf("templates/%s/%s/%s", kindFolder, langFolder, filename)
+	b, err := fs.ReadFile(tpls, filepath)
+	if err != nil {
+		log.Panic("read file failed: ", err)
+	}
+	return string(b)
+}
diff --git a/i18n/load_agents.go b/i18n/load_agents.go
new file mode 100644
index 000000000..cacae2dbc
--- /dev/null
+++ b/i18n/load_agents.go
@@ -0,0 +1,40 @@
+package i18n
+
+import (
+	"encoding/json"
+	"log"
+	"sync"
+)
+
+// AgentsMustLoad loads file of agents by the given language and filename.
+// Will panic if any problem occurs, including unsupported language,
+// unknown filename and other problems.
+func AgentsMustLoad(lang Lang, filename string) string {
+	return mustLoad(lang, "agents", filename)
+}
+
+// nolint:gochecknoglobals
+var langAgentsPhraseMap sync.Map
+
+// AgentsMustPhrase loads phrase of agents by the given language and key.
+// Will panic if any problem occurs, including unsupported language,
+// unknown key and other problems.
+func AgentsMustPhrase(lang Lang, key string) string {
+	var agentsPhraseMap map[string]string
+	valAny, ok := langAgentsPhraseMap.Load(lang)
+	if ok {
+		agentsPhraseMap, _ = valAny.(map[string]string)
+	} else {
+		s := AgentsMustLoad(lang, "phrase.json")
+		agentsPhraseMap = make(map[string]string)
+		if err := json.Unmarshal([]byte(s), &agentsPhraseMap); err != nil {
+			log.Panic("unmarshal phrase failed:", err)
+		}
+		langAgentsPhraseMap.Store(lang, agentsPhraseMap)
+	}
+	val, ok := agentsPhraseMap[key]
+	if !ok {
+		log.Panic("there is no such key in phrase:", key)
+	}
+	return val
+}
diff --git a/i18n/load_agents_test.go b/i18n/load_agents_test.go
new file mode 100644
index 000000000..8d1a3a2c4
--- /dev/null
+++ b/i18n/load_agents_test.go
@@ -0,0 +1,59 @@
+package i18n
+
+import (
+	"strings"
+	"testing"
+)
+
+func TestAgentsMustPhrase(t *testing.T) {
+	t.Parallel()
+	type args struct {
+		lang Lang
+		key  string
+	}
+	tests := []struct {
+		name          string
+		args          args
+		want          string
+		wantPanic     bool
+		wantPanicLike string
+	}{
+		{
+			name: "should succeed",
+			args: args{
+				lang: ZH,
+				key:  "thought",
+			},
+			want: "思考:",
+		},
+		{
+			name: "should panic",
+			args: args{
+				lang: ZH,
+				key:  "invalid key",
+			},
+			wantPanic:     true,
+			wantPanicLike: "there is no such key in phrase",
+		},
+	}
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			t.Parallel()
+			if tt.wantPanic {
+				defer func() {
+					if r := recover(); r == nil {
+						t.Errorf("want panic, but did not happen")
+					} else if s, ok := r.(string); !ok {
+						t.Errorf("unexpected panic type")
+					} else if !strings.Contains(s, tt.wantPanicLike) {
+						t.Errorf("panic = %v, want %v", r, tt.wantPanicLike)
+					}
+				}()
+			}
+
+			if got := AgentsMustPhrase(tt.args.lang, tt.args.key); got != tt.want {
+				t.Errorf("AgentsMustPhrase() = %v, want %v", got, tt.want)
+			}
+		})
+	}
+}
diff --git a/i18n/load_test.go b/i18n/load_test.go
new file mode 100644
index 000000000..9f2440288
--- /dev/null
+++ b/i18n/load_test.go
@@ -0,0 +1,81 @@
+package i18n
+
+import (
+	"strings"
+	"testing"
+)
+
+func Test_mustLoad(t *testing.T) {
+	t.Parallel()
+	type args struct {
+		lang       Lang
+		kindFolder string
+		filename   string
+	}
+	tests := []struct {
+		name          string
+		args          args
+		want          string
+		wantPanic     bool
+		wantPanicLike string
+	}{
+		{
+			name: "should succeed",
+			args: args{
+				lang:       ZH,
+				kindFolder: "agents",
+				filename:   "mrkl_prompt_format_instructions.txt",
+			},
+			want: `严格遵循以下格式:
+
+问题: 你要回答的问题
+思考: 针对这个问题，你的思考过程
+工具: 你将要使用的工具，只能在 [ {{.tool_names}} ] 中选择一个
+工具参数: 使用工具所需的参数
+结果: 使用工具获得的结果
+...（『思考/工具/工具参数/结果』可以重复多次）
+思考: 我知道最终答案了
+最终答案: 问题的最终答案`,
+		},
+		{
+			name: "should panic due to unknown language",
+			args: args{
+				lang:       Lang(-1),
+				kindFolder: "whatever",
+				filename:   "whatever",
+			},
+			wantPanic:     true,
+			wantPanicLike: "unknown language",
+		},
+		{
+			name: "should panic due to reading file failed",
+			args: args{
+				lang:       EN,
+				kindFolder: "agents",
+				filename:   "whatever",
+			},
+			wantPanic:     true,
+			wantPanicLike: "read file failed",
+		},
+	}
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			t.Parallel()
+			if tt.wantPanic {
+				defer func() {
+					if r := recover(); r == nil {
+						t.Errorf("want panic, but did not happen")
+					} else if s, ok := r.(string); !ok {
+						t.Errorf("unexpected panic type")
+					} else if !strings.Contains(s, tt.wantPanicLike) {
+						t.Errorf("panic = %v, want %v", r, tt.wantPanicLike)
+					}
+				}()
+			}
+
+			if got := mustLoad(tt.args.lang, tt.args.kindFolder, tt.args.filename); got != tt.want {
+				t.Errorf("mustLoad() = %v, want %v", got, tt.want)
+			}
+		})
+	}
+}
diff --git a/agents/prompts/conversational_format_instructions.txt b/i18n/templates/agents/en/conversational_prompt_format_instructions.txt
similarity index 100%
rename from agents/prompts/conversational_format_instructions.txt
rename to i18n/templates/agents/en/conversational_prompt_format_instructions.txt
diff --git a/agents/prompts/conversational_prefix.txt b/i18n/templates/agents/en/conversational_prompt_prefix.txt
similarity index 100%
rename from agents/prompts/conversational_prefix.txt
rename to i18n/templates/agents/en/conversational_prompt_prefix.txt
diff --git a/agents/prompts/conversational_suffix.txt b/i18n/templates/agents/en/conversational_prompt_suffix.txt
similarity index 100%
rename from agents/prompts/conversational_suffix.txt
rename to i18n/templates/agents/en/conversational_prompt_suffix.txt
diff --git a/i18n/templates/agents/en/mrkl_prompt_format_instructions.txt b/i18n/templates/agents/en/mrkl_prompt_format_instructions.txt
new file mode 100644
index 000000000..06a7c2995
--- /dev/null
+++ b/i18n/templates/agents/en/mrkl_prompt_format_instructions.txt
@@ -0,0 +1,10 @@
+Use the following format:
+
+Question: the input question you must answer
+Thought: you should always think about what to do
+Action: the action to take, should be one of [ {{.tool_names}} ]
+Action Input: the input to the action
+Observation: the result of the action
+... (this Thought/Action/Action Input/Observation can repeat N times)
+Thought: I now know the final answer
+Final Answer: the final answer to the original input question
\ No newline at end of file
diff --git a/i18n/templates/agents/en/mrkl_prompt_prefix.txt b/i18n/templates/agents/en/mrkl_prompt_prefix.txt
new file mode 100644
index 000000000..e35d626b9
--- /dev/null
+++ b/i18n/templates/agents/en/mrkl_prompt_prefix.txt
@@ -0,0 +1,4 @@
+Today is {{.today}}.
+Answer the following questions as best you can. You have access to the following tools:
+
+{{.tool_descriptions}}
\ No newline at end of file
diff --git a/i18n/templates/agents/en/mrkl_prompt_suffix.txt b/i18n/templates/agents/en/mrkl_prompt_suffix.txt
new file mode 100644
index 000000000..12235f61e
--- /dev/null
+++ b/i18n/templates/agents/en/mrkl_prompt_suffix.txt
@@ -0,0 +1,4 @@
+Begin!
+
+Question: {{.input}}
+Thought: {{.agent_scratchpad}}
\ No newline at end of file
diff --git a/i18n/templates/agents/en/phrase.json b/i18n/templates/agents/en/phrase.json
new file mode 100644
index 000000000..23648a6c7
--- /dev/null
+++ b/i18n/templates/agents/en/phrase.json
@@ -0,0 +1,11 @@
+{
+  "thought": "Thought:",
+  "action": "Action:",
+  "action input": "Action Input:",
+  "observation": "Observation:",
+  "output key": "Output:",
+  "system message": "You are a helpful AI assistant.",
+  "today format": "January 02, 2006",
+  "mrkl final answer": "Final Answer:",
+  "conversational final answer": "AI:"
+}
\ No newline at end of file
diff --git a/i18n/templates/agents/zh/conversational_prompt_format_instructions.txt b/i18n/templates/agents/zh/conversational_prompt_format_instructions.txt
new file mode 100644
index 000000000..c85241666
--- /dev/null
+++ b/i18n/templates/agents/zh/conversational_prompt_format_instructions.txt
@@ -0,0 +1,11 @@
+使用工具时，遵循以下格式:
+
+思考: 我需要使用工具吗？是的
+工具: 你将要使用的工具，只能在 [ {{.tool_names}} ] 中选择一个
+工具参数: 使用工具所需的参数
+结果: 使用工具获得的结果
+
+当你对人类有回应时，或者如果你不需要使用工具，你必须使用以下格式：
+
+思考: 我需要使用工具吗？不
+AI: [你的回答]
diff --git a/i18n/templates/agents/zh/conversational_prompt_prefix.txt b/i18n/templates/agents/zh/conversational_prompt_prefix.txt
new file mode 100644
index 000000000..a0937365a
--- /dev/null
+++ b/i18n/templates/agents/zh/conversational_prompt_prefix.txt
@@ -0,0 +1,6 @@
+工具:
+------
+
+你可以使用以下工具:
+
+{{.tool_descriptions}}
\ No newline at end of file
diff --git a/i18n/templates/agents/zh/conversational_prompt_suffix.txt b/i18n/templates/agents/zh/conversational_prompt_suffix.txt
new file mode 100644
index 000000000..c53222be4
--- /dev/null
+++ b/i18n/templates/agents/zh/conversational_prompt_suffix.txt
@@ -0,0 +1,8 @@
+开始吧！
+
+对话历史:
+{{.history}}
+
+最新输入: {{.input}}
+
+思考: {{.agent_scratchpad}}
\ No newline at end of file
diff --git a/i18n/templates/agents/zh/mrkl_prompt_format_instructions.txt b/i18n/templates/agents/zh/mrkl_prompt_format_instructions.txt
new file mode 100644
index 000000000..f27d63043
--- /dev/null
+++ b/i18n/templates/agents/zh/mrkl_prompt_format_instructions.txt
@@ -0,0 +1,10 @@
+严格遵循以下格式:
+
+问题: 你要回答的问题
+思考: 针对这个问题，你的思考过程
+工具: 你将要使用的工具，只能在 [ {{.tool_names}} ] 中选择一个
+工具参数: 使用工具所需的参数
+结果: 使用工具获得的结果
+...（『思考/工具/工具参数/结果』可以重复多次）
+思考: 我知道最终答案了
+最终答案: 问题的最终答案
\ No newline at end of file
diff --git a/i18n/templates/agents/zh/mrkl_prompt_prefix.txt b/i18n/templates/agents/zh/mrkl_prompt_prefix.txt
new file mode 100644
index 000000000..9330b199f
--- /dev/null
+++ b/i18n/templates/agents/zh/mrkl_prompt_prefix.txt
@@ -0,0 +1,4 @@
+今天是 {{.today}}。
+尽可能地回答问题。你可以使用以下工具:
+
+{{.tool_descriptions}}
\ No newline at end of file
diff --git a/i18n/templates/agents/zh/mrkl_prompt_suffix.txt b/i18n/templates/agents/zh/mrkl_prompt_suffix.txt
new file mode 100644
index 000000000..a10f46ee9
--- /dev/null
+++ b/i18n/templates/agents/zh/mrkl_prompt_suffix.txt
@@ -0,0 +1,4 @@
+开始吧！
+
+问题: {{.input}}
+思考: {{.agent_scratchpad}}
\ No newline at end of file
diff --git a/i18n/templates/agents/zh/phrase.json b/i18n/templates/agents/zh/phrase.json
new file mode 100644
index 000000000..463dc36e6
--- /dev/null
+++ b/i18n/templates/agents/zh/phrase.json
@@ -0,0 +1,11 @@
+{
+  "thought": "思考:",
+  "action": "工具:",
+  "action input": "工具参数:",
+  "observation": "结果:",
+  "output key": "输出:",
+  "system message": "你是一个牛逼的AI助手。",
+  "today format": "2006年01月02日",
+  "mrkl final answer": "最终答案:",
+  "conversational final answer": "AI:"
+}
\ No newline at end of file