Skip to content

Database governance built for humans and agents — controlling changes and access across every major database.

feat: track page agent thread token usage (#19697)

p0nydad4b296a2df
diff --git a/backend/api/mcp/gen/openapi.yaml b/backend/api/mcp/gen/openapi.yaml
index ecc3a842dd..f3a9e2642d 100644
--- a/backend/api/mcp/gen/openapi.yaml
+++ b/backend/api/mcp/gen/openapi.yaml
@@ -7162,6 +7162,12 @@ components:
$ref: '#/components/schemas/bytebase.v1.AIChatToolCall'
title: tool_calls
description: Tool calls the AI wants to make.
+ usage:
+ oneOf:
+ - $ref: '#/components/schemas/bytebase.v1.AIChatUsage'
+ - type: "null"
+ title: usage
+ description: Token usage for this provider call, when available.
title: AIChatResponse
additionalProperties: false
description: Response message for AIService.Chat.
@@ -7209,6 +7215,17 @@ components:
title: AIChatToolDefinition
additionalProperties: false
description: A tool definition that the AI can invoke.
+ bytebase.v1.AIChatUsage:
+ type: object
+ properties:
+ totalTokens:
+ type: integer
+ title: total_tokens
+ format: int32
+ description: Total tokens used by the provider call.
+ title: AIChatUsage
+ additionalProperties: false
+ description: Token usage for a single AI provider call.
bytebase.v1.AICompletionRequest:
type: object
properties:
diff --git a/backend/api/v1/ai_service.go b/backend/api/v1/ai_service.go
index 0c2f1044eb..62609f1f60 100644
--- a/backend/api/v1/ai_service.go
+++ b/backend/api/v1/ai_service.go
@@ -91,6 +91,10 @@ type chatOpenAIFunctionCallRef struct {
Arguments string `json:"arguments"`
}
+type chatOpenAIUsage struct {
+ TotalTokens int32 `json:"total_tokens"`
+}
+
type chatOpenAIResponse struct {
Choices []struct {
Message struct {
@@ -99,6 +103,7 @@ type chatOpenAIResponse struct {
ToolCalls []json.RawMessage `json:"tool_calls"`
} `json:"message"`
} `json:"choices"`
+ Usage *chatOpenAIUsage `json:"usage,omitempty"`
}
// chatOpenAIToolCallParsed extracts the fields we need from a raw tool call JSON.
@@ -276,6 +281,9 @@ func (*AIService) chatOpenAI(ctx context.Context, aiSetting *storepb.AISetting,
}
result := &v1pb.AIChatResponse{}
+ if resp.Usage != nil {
+ result.Usage = newAIChatUsage(resp.Usage.TotalTokens)
+ }
if len(resp.Choices) > 0 {
msg := resp.Choices[0].Message
result.Content = msg.Content
@@ -333,8 +341,14 @@ type chatClaudeContentBlock struct {
Content string `json:"content,omitempty"`
}
+type chatClaudeUsage struct {
+ InputTokens int32 `json:"input_tokens"`
+ OutputTokens int32 `json:"output_tokens"`
+}
+
type chatClaudeResponse struct {
Content []chatClaudeContentBlock `json:"content"`
+ Usage *chatClaudeUsage `json:"usage,omitempty"`
}
func (*AIService) chatClaude(ctx context.Context, aiSetting *storepb.AISetting, request *v1pb.AIChatRequest) (*connect.Response[v1pb.AIChatResponse], error) {
@@ -434,6 +448,9 @@ func (*AIService) chatClaude(ctx context.Context, aiSetting *storepb.AISetting,
}
result := &v1pb.AIChatResponse{}
+ if resp.Usage != nil {
+ result.Usage = newAIChatUsage(resp.Usage.InputTokens + resp.Usage.OutputTokens)
+ }
var textContent string
for _, block := range resp.Content {
switch block.Type {
@@ -497,12 +514,17 @@ type chatGeminiFunctionDecl struct {
Parameters any `json:"parameters"`
}
+type chatGeminiUsage struct {
+ TotalTokenCount int32 `json:"totalTokenCount"`
+}
+
type chatGeminiResponse struct {
Candidates []struct {
Content struct {
Parts []chatGeminiResponsePart `json:"parts"`
} `json:"content"`
} `json:"candidates"`
+ UsageMetadata *chatGeminiUsage `json:"usageMetadata,omitempty"`
}
type chatGeminiResponsePart struct {
@@ -618,6 +640,9 @@ func (*AIService) chatGemini(ctx context.Context, aiSetting *storepb.AISetting,
}
result := &v1pb.AIChatResponse{}
+ if resp.UsageMetadata != nil {
+ result.Usage = newAIChatUsage(resp.UsageMetadata.TotalTokenCount)
+ }
if len(resp.Candidates) > 0 {
var textContent string
for _, part := range resp.Candidates[0].Content.Parts {
@@ -651,6 +676,15 @@ func (*AIService) chatGemini(ctx context.Context, aiSetting *storepb.AISetting,
return connect.NewResponse(result), nil
}
+func newAIChatUsage(totalTokens int32) *v1pb.AIChatUsage {
+ if totalTokens <= 0 {
+ return nil
+ }
+ return &v1pb.AIChatUsage{
+ TotalTokens: totalTokens,
+ }
+}
+
// doHTTPRequest executes an HTTP request and returns the response body.
func doHTTPRequest(_ context.Context, httpReq *http.Request, provider string) ([]byte, error) {
client := &http.Client{}
diff --git a/backend/generated-go/v1/ai_service.pb.go b/backend/generated-go/v1/ai_service.pb.go
index 2d70c757be..3fdc64e964 100644
--- a/backend/generated-go/v1/ai_service.pb.go
+++ b/backend/generated-go/v1/ai_service.pb.go
@@ -349,20 +349,68 @@ func (x *AIChatRequest) GetToolDefinitions() []*AIChatToolDefinition {
return nil
}
+// Token usage for a single AI provider call.
+type AIChatUsage struct {
+ state protoimpl.MessageState `protogen:"open.v1"`
+ // Total tokens used by the provider call.
+ TotalTokens int32 `protobuf:"varint,1,opt,name=total_tokens,json=totalTokens,proto3" json:"total_tokens,omitempty"`
+ unknownFields protoimpl.UnknownFields
+ sizeCache protoimpl.SizeCache
+}
+
+func (x *AIChatUsage) Reset() {
+ *x = AIChatUsage{}
+ mi := &file_v1_ai_service_proto_msgTypes[4]
+ ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
+ ms.StoreMessageInfo(mi)
+}
+
+func (x *AIChatUsage) String() string {
+ return protoimpl.X.MessageStringOf(x)
+}
+
+func (*AIChatUsage) ProtoMessage() {}
+
+func (x *AIChatUsage) ProtoReflect() protoreflect.Message {
+ mi := &file_v1_ai_service_proto_msgTypes[4]
+ if x != nil {
+ ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
+ if ms.LoadMessageInfo() == nil {
+ ms.StoreMessageInfo(mi)
+ }
+ return ms
+ }
+ return mi.MessageOf(x)
+}
+
+// Deprecated: Use AIChatUsage.ProtoReflect.Descriptor instead.
+func (*AIChatUsage) Descriptor() ([]byte, []int) {
+ return file_v1_ai_service_proto_rawDescGZIP(), []int{4}
+}
+
+func (x *AIChatUsage) GetTotalTokens() int32 {
+ if x != nil {
+ return x.TotalTokens
+ }
+ return 0
+}
+
// Response message for AIService.Chat.
type AIChatResponse struct {
state protoimpl.MessageState `protogen:"open.v1"`
// The text content of the AI response. Optional when the response only contains tool calls.
Content *string `protobuf:"bytes,1,opt,name=content,proto3,oneof" json:"content,omitempty"`
// Tool calls the AI wants to make.
- ToolCalls []*AIChatToolCall `protobuf:"bytes,2,rep,name=tool_calls,json=toolCalls,proto3" json:"tool_calls,omitempty"`
+ ToolCalls []*AIChatToolCall `protobuf:"bytes,2,rep,name=tool_calls,json=toolCalls,proto3" json:"tool_calls,omitempty"`
+ // Token usage for this provider call, when available.
+ Usage *AIChatUsage `protobuf:"bytes,3,opt,name=usage,proto3,oneof" json:"usage,omitempty"`
unknownFields protoimpl.UnknownFields
sizeCache protoimpl.SizeCache
}
func (x *AIChatResponse) Reset() {
*x = AIChatResponse{}
- mi := &file_v1_ai_service_proto_msgTypes[4]
+ mi := &file_v1_ai_service_proto_msgTypes[5]
ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
ms.StoreMessageInfo(mi)
}
@@ -374,7 +422,7 @@ func (x *AIChatResponse) String() string {
func (*AIChatResponse) ProtoMessage() {}
func (x *AIChatResponse) ProtoReflect() protoreflect.Message {
- mi := &file_v1_ai_service_proto_msgTypes[4]
+ mi := &file_v1_ai_service_proto_msgTypes[5]
if x != nil {
ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
if ms.LoadMessageInfo() == nil {
@@ -387,7 +435,7 @@ func (x *AIChatResponse) ProtoReflect() protoreflect.Message {
// Deprecated: Use AIChatResponse.ProtoReflect.Descriptor instead.
func (*AIChatResponse) Descriptor() ([]byte, []int) {
- return file_v1_ai_service_proto_rawDescGZIP(), []int{4}
+ return file_v1_ai_service_proto_rawDescGZIP(), []int{5}
}
func (x *AIChatResponse) GetContent() string {
@@ -404,6 +452,13 @@ func (x *AIChatResponse) GetToolCalls() []*AIChatToolCall {
return nil
}
+func (x *AIChatResponse) GetUsage() *AIChatUsage {
+ if x != nil {
+ return x.Usage
+ }
+ return nil
+}
+
var File_v1_ai_service_proto protoreflect.FileDescriptor
const file_v1_ai_service_proto_rawDesc = "" +
@@ -431,13 +486,17 @@ const file_v1_ai_service_proto_rawDesc = "" +
"\t_metadata\"\x95\x01\n" +
"\rAIChatRequest\x126\n" +
"\bmessages\x18\x01 \x03(\v2\x1a.bytebase.v1.AIChatMessageR\bmessages\x12L\n" +
- "\x10tool_definitions\x18\x02 \x03(\v2!.bytebase.v1.AIChatToolDefinitionR\x0ftoolDefinitions\"w\n" +
+ "\x10tool_definitions\x18\x02 \x03(\v2!.bytebase.v1.AIChatToolDefinitionR\x0ftoolDefinitions\"0\n" +
+ "\vAIChatUsage\x12!\n" +
+ "\ftotal_tokens\x18\x01 \x01(\x05R\vtotalTokens\"\xb6\x01\n" +
"\x0eAIChatResponse\x12\x1d\n" +
"\acontent\x18\x01 \x01(\tH\x00R\acontent\x88\x01\x01\x12:\n" +
"\n" +
- "tool_calls\x18\x02 \x03(\v2\x1b.bytebase.v1.AIChatToolCallR\ttoolCallsB\n" +
+ "tool_calls\x18\x02 \x03(\v2\x1b.bytebase.v1.AIChatToolCallR\ttoolCalls\x123\n" +
+ "\x05usage\x18\x03 \x01(\v2\x18.bytebase.v1.AIChatUsageH\x01R\x05usage\x88\x01\x01B\n" +
"\n" +
- "\b_content*\xbc\x01\n" +
+ "\b_contentB\b\n" +
+ "\x06_usage*\xbc\x01\n" +
"\x11AIChatMessageRole\x12$\n" +
" AI_CHAT_MESSAGE_ROLE_UNSPECIFIED\x10\x00\x12\x1f\n" +
"\x1bAI_CHAT_MESSAGE_ROLE_SYSTEM\x10\x01\x12\x1d\n" +
@@ -461,14 +520,15 @@ func file_v1_ai_service_proto_rawDescGZIP() []byte {
}
var file_v1_ai_service_proto_enumTypes = make([]protoimpl.EnumInfo, 1)
-var file_v1_ai_service_proto_msgTypes = make([]protoimpl.MessageInfo, 5)
+var file_v1_ai_service_proto_msgTypes = make([]protoimpl.MessageInfo, 6)
var file_v1_ai_service_proto_goTypes = []any{
(AIChatMessageRole)(0), // 0: bytebase.v1.AIChatMessageRole
(*AIChatMessage)(nil), // 1: bytebase.v1.AIChatMessage
(*AIChatToolDefinition)(nil), // 2: bytebase.v1.AIChatToolDefinition
(*AIChatToolCall)(nil), // 3: bytebase.v1.AIChatToolCall
(*AIChatRequest)(nil), // 4: bytebase.v1.AIChatRequest
- (*AIChatResponse)(nil), // 5: bytebase.v1.AIChatResponse
+ (*AIChatUsage)(nil), // 5: bytebase.v1.AIChatUsage
+ (*AIChatResponse)(nil), // 6: bytebase.v1.AIChatResponse
}
var file_v1_ai_service_proto_depIdxs = []int32{
0, // 0: bytebase.v1.AIChatMessage.role:type_name -> bytebase.v1.AIChatMessageRole
@@ -476,13 +536,14 @@ var file_v1_ai_service_proto_depIdxs = []int32{
1, // 2: bytebase.v1.AIChatRequest.messages:type_name -> bytebase.v1.AIChatMessage
2, // 3: bytebase.v1.AIChatRequest.tool_definitions:type_name -> bytebase.v1.AIChatToolDefinition
3, // 4: bytebase.v1.AIChatResponse.tool_calls:type_name -> bytebase.v1.AIChatToolCall
- 4, // 5: bytebase.v1.AIService.Chat:input_type -> bytebase.v1.AIChatRequest
- 5, // 6: bytebase.v1.AIService.Chat:output_type -> bytebase.v1.AIChatResponse
- 6, // [6:7] is the sub-list for method output_type
- 5, // [5:6] is the sub-list for method input_type
- 5, // [5:5] is the sub-list for extension type_name
- 5, // [5:5] is the sub-list for extension extendee
- 0, // [0:5] is the sub-list for field type_name
+ 5, // 5: bytebase.v1.AIChatResponse.usage:type_name -> bytebase.v1.AIChatUsage
+ 4, // 6: bytebase.v1.AIService.Chat:input_type -> bytebase.v1.AIChatRequest
+ 6, // 7: bytebase.v1.AIService.Chat:output_type -> bytebase.v1.AIChatResponse
+ 7, // [7:8] is the sub-list for method output_type
+ 6, // [6:7] is the sub-list for method input_type
+ 6, // [6:6] is the sub-list for extension type_name
+ 6, // [6:6] is the sub-list for extension extendee
+ 0, // [0:6] is the sub-list for field type_name
}
func init() { file_v1_ai_service_proto_init() }
@@ -493,14 +554,14 @@ func file_v1_ai_service_proto_init() {
file_v1_annotation_proto_init()
file_v1_ai_service_proto_msgTypes[0].OneofWrappers = []any{}
file_v1_ai_service_proto_msgTypes[2].OneofWrappers = []any{}
- file_v1_ai_service_proto_msgTypes[4].OneofWrappers = []any{}
+ file_v1_ai_service_proto_msgTypes[5].OneofWrappers = []any{}
type x struct{}
out := protoimpl.TypeBuilder{
File: protoimpl.DescBuilder{
GoPackagePath: reflect.TypeOf(x{}).PkgPath(),
RawDescriptor: unsafe.Slice(unsafe.StringData(file_v1_ai_service_proto_rawDesc), len(file_v1_ai_service_proto_rawDesc)),
NumEnums: 1,
- NumMessages: 5,
+ NumMessages: 6,
NumExtensions: 0,
NumServices: 1,
},
diff --git a/backend/generated-go/v1/ai_service_equal.pb.go b/backend/generated-go/v1/ai_service_equal.pb.go
index 6e7a24606d..d341601f49 100644
--- a/backend/generated-go/v1/ai_service_equal.pb.go
+++ b/backend/generated-go/v1/ai_service_equal.pb.go
@@ -97,6 +97,19 @@ func (x *AIChatRequest) Equal(y *AIChatRequest) bool {
return true
}
+func (x *AIChatUsage) Equal(y *AIChatUsage) bool {
+ if x == y {
+ return true
+ }
+ if x == nil || y == nil {
+ return x == nil && y == nil
+ }
+ if x.TotalTokens != y.TotalTokens {
+ return false
+ }
+ return true
+}
+
func (x *AIChatResponse) Equal(y *AIChatResponse) bool {
if x == y {
return true
@@ -115,5 +128,8 @@ func (x *AIChatResponse) Equal(y *AIChatResponse) bool {
return false
}
}
+ if !x.Usage.Equal(y.Usage) {
+ return false
+ }
return true
}
diff --git a/frontend/src/locales/en-US.json b/frontend/src/locales/en-US.json
index 7d4be9ff30..436ce4c49a 100644
--- a/frontend/src/locales/en-US.json
+++ b/frontend/src/locales/en-US.json
@@ -2982,6 +2982,7 @@
"thread-status-awaiting-user": "Awaiting input",
"thread-status-error": "Error",
"thread-status-interrupted": "Interrupted",
+ "thread-total-tokens": "Tokens used: {count}",
"reply": "Reply",
"confirm": "Confirm",
"cancel": "Cancel",
diff --git a/frontend/src/locales/es-ES.json b/frontend/src/locales/es-ES.json
index a552028640..91ceff5f20 100644
--- a/frontend/src/locales/es-ES.json
+++ b/frontend/src/locales/es-ES.json
@@ -2491,6 +2491,7 @@
"thread-status-awaiting-user": "Esperando entrada",
"thread-status-error": "Error",
"thread-status-interrupted": "Interrumpido",
+ "thread-total-tokens": "Tokens usados: {count}",
"reply": "Responder",
"confirm": "Confirmar",
"cancel": "Cancelar",
diff --git a/frontend/src/locales/ja-JP.json b/frontend/src/locales/ja-JP.json
index 01f6a16195..01493b1b1a 100644
--- a/frontend/src/locales/ja-JP.json
+++ b/frontend/src/locales/ja-JP.json
@@ -2491,6 +2491,7 @@
"thread-status-awaiting-user": "入力待ち",
"thread-status-error": "エラー",
"thread-status-interrupted": "中断",
+ "thread-total-tokens": "使用済みトークン: {count}",
"reply": "返信",
"confirm": "確認",
"cancel": "キャンセル",
diff --git a/frontend/src/locales/vi-VN.json b/frontend/src/locales/vi-VN.json
index 6b269e7011..28b29fb8ca 100644
--- a/frontend/src/locales/vi-VN.json
+++ b/frontend/src/locales/vi-VN.json
@@ -2491,6 +2491,7 @@
"thread-status-awaiting-user": "Đang chờ nhập",
"thread-status-error": "Lỗi",
"thread-status-interrupted": "Bị gián đoạn",
+ "thread-total-tokens": "Token đã dùng: {count}",
"reply": "Trả lời",
"confirm": "Xác nhận",
"cancel": "Hủy",
diff --git a/frontend/src/locales/zh-CN.json b/frontend/src/locales/zh-CN.json
index 630f5e9a0f..35a8292286 100644
--- a/frontend/src/locales/zh-CN.json
+++ b/frontend/src/locales/zh-CN.json
@@ -2982,6 +2982,7 @@
"thread-status-awaiting-user": "等待输入",
"thread-status-error": "错误",
"thread-status-interrupted": "已中断",
+ "thread-total-tokens": "已用令牌:{count}",
"reply": "回复",
"confirm": "确认",
"cancel": "取消",
diff --git a/frontend/src/plugins/agent/components/AgentInput.test.ts b/frontend/src/plugins/agent/components/AgentInput.test.ts
index 1733d4eb27..7f150a3159 100644
--- a/frontend/src/plugins/agent/components/AgentInput.test.ts
+++ b/frontend/src/plugins/agent/components/AgentInput.test.ts
@@ -224,6 +224,31 @@ describe("AgentInput", () => {
});
});
+ test("increments the current thread token total from loop usage", async () => {
+ const store = useAgentStore();
+ const threadId = store.currentThreadId!;
+
+ mockRunAgentLoop.mockResolvedValue({
+ kind: "completed",
+ text: "Done",
+ success: true,
+ explicit: true,
+ totalTokensUsed: 123,
+ });
+
+ const wrapper = mount(AgentInput, {
+ global: {
+ plugins: [pinia, i18n],
+ },
+ });
+
+ await wrapper.find("textarea").setValue("inspect this page");
+ await findButtonByText(wrapper, "Send")!.trigger("click");
+ await flushPromises();
+
+ expect(store.getThread(threadId)?.totalTokensUsed).toBe(123);
+ });
+
test("refreshes the saved thread page snapshot after navigate before a follow-up turn", async () => {
const store = useAgentStore();
const threadId = store.currentThreadId!;
diff --git a/frontend/src/plugins/agent/components/AgentInput.vue b/frontend/src/plugins/agent/components/AgentInput.vue
index 87ae491a55..1b12b9752e 100644
--- a/frontend/src/plugins/agent/components/AgentInput.vue
+++ b/frontend/src/plugins/agent/components/AgentInput.vue
@@ -188,6 +188,11 @@ async function runThread(
controller.signal
);
+ agentStore.incrementThreadTotalTokens(
+ threadId,
+ outcome.totalTokensUsed ?? 0
+ );
+
if (runToken !== currentRunToken) {
return;
}
diff --git a/frontend/src/plugins/agent/components/AgentWindow.vue b/frontend/src/plugins/agent/components/AgentWindow.vue
index d0d5822ba0..bc0a4ef3d0 100644
--- a/frontend/src/plugins/agent/components/AgentWindow.vue
+++ b/frontend/src/plugins/agent/components/AgentWindow.vue
@@ -110,6 +110,15 @@ const currentThreadStatusClass = computed(() => {
return "bg-amber-50 text-amber-600";
});
+const tokenFormatter = new Intl.NumberFormat();
+const currentThreadTokenUsageLabel = computed(() =>
+ t("agent.thread-total-tokens", {
+ count: tokenFormatter.format(
+ agentStore.currentThread?.totalTokensUsed ?? 0
+ ),
+ })
+);
+
function syncSize(width: number, height: number) {
const size = getDisplaySize(width, height);
agentStore.size.width = size.width;
@@ -336,6 +345,9 @@ onBeforeUnmount(() => {
>
{{ currentThreadStatusLabel }}
</span>
+ <span class="truncate text-xs text-gray-500">
+ {{ currentThreadTokenUsageLabel }}
+ </span>
</div>
<div class="flex items-center gap-x-1">
<button
diff --git a/frontend/src/plugins/agent/logic/agentLoop.test.ts b/frontend/src/plugins/agent/logic/agentLoop.test.ts
index ab6f2d5f80..8b12f0b5fc 100644
--- a/frontend/src/plugins/agent/logic/agentLoop.test.ts
+++ b/frontend/src/plugins/agent/logic/agentLoop.test.ts
@@ -54,6 +54,7 @@ describe("runAgentLoop", () => {
text: "Finished successfully",
success: true,
explicit: true,
+ totalTokensUsed: 0,
});
expect(onToolResult).toHaveBeenCalledWith(
"tool-1",
@@ -80,6 +81,50 @@ describe("runAgentLoop", () => {
});
});
+ test("accumulates token usage across repeated provider calls", async () => {
+ vi.mocked(aiServiceClientConnect.chat)
+ .mockResolvedValueOnce({
+ content: "",
+ toolCalls: [
+ {
+ id: "tool-1",
+ name: "search_api",
+ arguments: JSON.stringify({ service: "SQLService" }),
+ metadata: "",
+ },
+ ],
+ usage: {
+ totalTokens: 11,
+ },
+ } as never)
+ .mockResolvedValueOnce({
+ content: "Final answer",
+ toolCalls: [],
+ usage: {
+ totalTokens: 7,
+ },
+ } as never);
+
+ const executeTool: ToolExecutor = vi.fn(async () => ({
+ kind: "tool_result" as const,
+ result: JSON.stringify({ ok: true }),
+ }));
+
+ const outcome = await runAgentLoop(
+ [{ role: "system", content: "system" }],
+ [],
+ executeTool
+ );
+
+ expect(outcome).toEqual({
+ kind: "completed",
+ text: "Final answer",
+ success: true,
+ explicit: false,
+ totalTokensUsed: 18,
+ });
+ });
+
test("replays completed tool flows with matched tool results", async () => {
vi.mocked(aiServiceClientConnect.chat)
.mockResolvedValueOnce({
@@ -195,6 +240,7 @@ describe("runAgentLoop", () => {
text: "Here is the answer",
success: true,
explicit: false,
+ totalTokensUsed: 0,
});
expect(onText).toHaveBeenCalledWith("Here is the answer");
expect(executeTool).not.toHaveBeenCalled();
@@ -254,6 +300,7 @@ describe("runAgentLoop", () => {
{ label: "Staging", value: "staging" },
],
},
+ totalTokensUsed: 0,
});
expect(onToolResult).not.toHaveBeenCalled();
});
@@ -312,6 +359,7 @@ describe("runAgentLoop", () => {
prompt: "Which environment should I use?",
kind: "input",
},
+ totalTokensUsed: 0,
});
expect(executeTool).toHaveBeenCalledTimes(1);
expect(executeTool).toHaveBeenCalledWith(
@@ -384,6 +432,7 @@ describe("runAgentLoop", () => {
text: "Finished successfully",
success: true,
explicit: true,
+ totalTokensUsed: 0,
});
expect(executeTool).toHaveBeenCalledTimes(1);
expect(executeTool).toHaveBeenCalledWith(
diff --git a/frontend/src/plugins/agent/logic/agentLoop.ts b/frontend/src/plugins/agent/logic/agentLoop.ts
index 53f72df407..d3d15379e9 100644
--- a/frontend/src/plugins/agent/logic/agentLoop.ts
+++ b/frontend/src/plugins/agent/logic/agentLoop.ts
@@ -120,11 +120,12 @@ export async function runAgentLoop(
): Promise<AgentLoopOutcome> {
const conversation: Message[] = [...messages];
const protoTools = tools.map(toolDefToProto);
+ let totalTokensUsed = 0;
try {
for (let i = 0; i < MAX_ITERATIONS; i++) {
if (signal?.aborted) {
- return { kind: "aborted" };
+ return { kind: "aborted", totalTokensUsed };
}
const protoMessages = conversation.map(messageToProto);
@@ -141,6 +142,8 @@ export async function runAgentLoop(
signal
);
+ totalTokensUsed += response.usage?.totalTokens ?? 0;
+
if (response.toolCalls.length > 0) {
const toolCalls: ToolCall[] = response.toolCalls.map((tc) => ({
id: tc.id,
@@ -206,6 +209,7 @@ export async function runAgentLoop(
outcome: {
kind: "awaiting_user",
ask: executionResult.ask,
+ totalTokensUsed,
},
blockedBy: {
toolCallId: tc.id,
@@ -231,6 +235,7 @@ export async function runAgentLoop(
text: executionResult.text,
success: executionResult.success,
explicit: true,
+ totalTokensUsed,
},
blockedBy: {
toolCallId: tc.id,
@@ -256,21 +261,24 @@ export async function runAgentLoop(
text,
success: true,
explicit: false,
+ totalTokensUsed,
};
}
return {
kind: "error",
error: new Error("Agent loop exceeded maximum iterations"),
+ totalTokensUsed,
};
} catch (err) {
const error = err instanceof Error ? err : new Error(String(err));
if (error.name === "AbortError") {
- return { kind: "aborted" };
+ return { kind: "aborted", totalTokensUsed };
}
return {
kind: "error",
error,
+ totalTokensUsed,
};
}
}
diff --git a/frontend/src/plugins/agent/logic/types.ts b/frontend/src/plugins/agent/logic/types.ts
index 446af59d94..6e44dc716d 100644
--- a/frontend/src/plugins/agent/logic/types.ts
+++ b/frontend/src/plugins/agent/logic/types.ts
@@ -46,6 +46,7 @@ export interface AgentThread {
createdTs: number;
updatedTs: number;
status: AgentThreadStatus;
+ totalTokensUsed: number;
page?: AgentThreadSnapshot;
lastError?: string | null;
interrupted?: boolean;
@@ -101,24 +102,28 @@ export type ToolExecutionResult =
ask: AgentPendingAsk;
};
+interface AgentLoopUsage {
+ totalTokensUsed?: number;
+}
+
export type AgentLoopOutcome =
- | {
+ | ({
kind: "completed";
text: string;
success: boolean;
explicit: boolean;
- }
- | {
+ } & AgentLoopUsage)
+ | ({
kind: "awaiting_user";
ask: AgentPendingAsk;
- }
- | {
+ } & AgentLoopUsage)
+ | ({
kind: "aborted";
- }
- | {
+ } & AgentLoopUsage)
+ | ({
kind: "error";
error: Error;
- };
+ } & AgentLoopUsage);
export type ToolExecutor = (
name: string,
diff --git a/frontend/src/plugins/agent/store/agent.test.ts b/frontend/src/plugins/agent/store/agent.test.ts
index 88cf610724..60ce94e72a 100644
--- a/frontend/src/plugins/agent/store/agent.test.ts
+++ b/frontend/src/plugins/agent/store/agent.test.ts
@@ -54,6 +54,7 @@ describe("useAgentStore", () => {
expect(store.threads).toHaveLength(1);
expect(store.currentThreadId).toBe(store.threads[0].id);
expect(store.messages).toEqual([]);
+ expect(store.threads[0].totalTokensUsed).toBe(0);
});
test("migrates legacy flat messages into a thread-aware state", () => {
@@ -179,6 +180,24 @@ describe("useAgentStore", () => {
});
});
+ test("increments and resets thread token totals", async () => {
+ const store = createStore();
+ const threadId = store.currentThreadId!;
+
+ store.incrementThreadTotalTokens(threadId, 120);
+ store.incrementThreadTotalTokens(threadId, 30);
+
+ expect(store.getThread(threadId)?.totalTokensUsed).toBe(150);
+
+ await nextTick();
+
+ const rehydratedStore = createStore();
+ expect(rehydratedStore.getThread(threadId)?.totalTokensUsed).toBe(150);
+
+ rehydratedStore.clearConversation(threadId);
+ expect(rehydratedStore.getThread(threadId)?.totalTokensUsed).toBe(0);
+ });
+
test("preserves an existing thread page when resuming a run", () => {
const store = createStore();
const threadId = store.currentThreadId!;
diff --git a/frontend/src/plugins/agent/store/agent.ts b/frontend/src/plugins/agent/store/agent.ts
index b8eeaf73da..ec2e2d6086 100644
--- a/frontend/src/plugins/agent/store/agent.ts
+++ b/frontend/src/plugins/agent/store/agent.ts
@@ -45,6 +45,7 @@ const createThreadRecord = (options: CreateThreadOptions = {}): AgentThread => {
createdTs: now,
updatedTs: now,
status: DEFAULT_THREAD_STATUS,
+ totalTokensUsed: 0,
page: options.page,
lastError: null,
interrupted: false,
@@ -174,6 +175,12 @@ const normalizeThread = (raw: unknown): AgentThread => {
createdTs: typeof thread.createdTs === "number" ? thread.createdTs : now,
updatedTs: typeof thread.updatedTs === "number" ? thread.updatedTs : now,
status,
+ totalTokensUsed:
+ typeof thread.totalTokensUsed === "number" &&
+ Number.isFinite(thread.totalTokensUsed) &&
+ thread.totalTokensUsed >= 0
+ ? thread.totalTokensUsed
+ : 0,
page:
isRecord(thread.page) &&
typeof thread.page.path === "string" &&
@@ -512,6 +519,19 @@ export const useAgentStore = defineStore("agent", () => {
return message;
};
+ const incrementThreadTotalTokens = (
+ threadId: string,
+ totalTokensUsed: number
+ ) => {
+ const thread = getThread(threadId);
+ if (!thread || totalTokensUsed <= 0) {
+ return null;
+ }
+ thread.totalTokensUsed += totalTokensUsed;
+ touchThread(threadId);
+ return thread;
+ };
+
const awaitUser = (threadId: string, pendingAsk: AgentPendingAsk) => {
setThreadStatus(threadId, "awaiting_user");
return setPendingAsk(threadId, pendingAsk);
@@ -547,6 +567,7 @@ export const useAgentStore = defineStore("agent", () => {
thread.title = "";
thread.page = undefined;
thread.status = DEFAULT_THREAD_STATUS;
+ thread.totalTokensUsed = 0;
thread.lastError = null;
thread.interrupted = false;
thread.runId = null;
@@ -760,6 +781,7 @@ export const useAgentStore = defineStore("agent", () => {
awaitUser,
answerPendingAsk,
touchThread,
+ incrementThreadTotalTokens,
addMessage,
removeMessagesByRunId,
appendToolCall,
diff --git a/frontend/src/types/proto-es/v1/ai_service_pb.d.ts b/frontend/src/types/proto-es/v1/ai_service_pb.d.ts
index 42090dca63..7df5b02d76 100644
--- a/frontend/src/types/proto-es/v1/ai_service_pb.d.ts
+++ b/frontend/src/types/proto-es/v1/ai_service_pb.d.ts
@@ -154,6 +154,26 @@ export declare type AIChatRequest = Message<"bytebase.v1.AIChatRequest"> & {
*/
export declare const AIChatRequestSchema: GenMessage<AIChatRequest>;
+/**
+ * Token usage for a single AI provider call.
+ *
+ * @generated from message bytebase.v1.AIChatUsage
+ */
+export declare type AIChatUsage = Message<"bytebase.v1.AIChatUsage"> & {
+ /**
+ * Total tokens used by the provider call.
+ *
+ * @generated from field: int32 total_tokens = 1;
+ */
+ totalTokens: number;
+};
+
+/**
+ * Describes the message bytebase.v1.AIChatUsage.
+ * Use `create(AIChatUsageSchema)` to create a new message.
+ */
+export declare const AIChatUsageSchema: GenMessage<AIChatUsage>;
+
/**
* Response message for AIService.Chat.
*
@@ -173,6 +193,13 @@ export declare type AIChatResponse = Message<"bytebase.v1.AIChatResponse"> & {
* @generated from field: repeated bytebase.v1.AIChatToolCall tool_calls = 2;
*/
toolCalls: AIChatToolCall[];
+
+ /**
+ * Token usage for this provider call, when available.
+ *
+ * @generated from field: optional bytebase.v1.AIChatUsage usage = 3;
+ */
+ usage?: AIChatUsage;
};
/**
diff --git a/frontend/src/types/proto-es/v1/ai_service_pb.js b/frontend/src/types/proto-es/v1/ai_service_pb.js
index 7e80684338..c6a33c1d44 100644
--- a/frontend/src/types/proto-es/v1/ai_service_pb.js
+++ b/frontend/src/types/proto-es/v1/ai_service_pb.js
@@ -10,7 +10,7 @@ import { file_v1_annotation } from "./annotation_pb";
* Describes the file v1/ai_service.proto.
*/
export const file_v1_ai_service = /*@__PURE__*/
- fileDesc("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", [file_google_api_annotations, file_v1_annotation]);
+ fileDesc("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", [file_google_api_annotations, file_v1_annotation]);
/**
* Describes the message bytebase.v1.AIChatMessage.
@@ -40,12 +40,19 @@ export const AIChatToolCallSchema = /*@__PURE__*/
export const AIChatRequestSchema = /*@__PURE__*/
messageDesc(file_v1_ai_service, 3);
+/**
+ * Describes the message bytebase.v1.AIChatUsage.
+ * Use `create(AIChatUsageSchema)` to create a new message.
+ */
+export const AIChatUsageSchema = /*@__PURE__*/
+ messageDesc(file_v1_ai_service, 4);
+
/**
* Describes the message bytebase.v1.AIChatResponse.
* Use `create(AIChatResponseSchema)` to create a new message.
*/
export const AIChatResponseSchema = /*@__PURE__*/
- messageDesc(file_v1_ai_service, 4);
+ messageDesc(file_v1_ai_service, 5);
/**
* Describes the enum bytebase.v1.AIChatMessageRole.
diff --git a/proto/gen/grpc-doc/openapi.yaml b/proto/gen/grpc-doc/openapi.yaml
index f185874a3c..11ce8208ba 100644
--- a/proto/gen/grpc-doc/openapi.yaml
+++ b/proto/gen/grpc-doc/openapi.yaml
@@ -16211,6 +16211,12 @@ components:
$ref: '#/components/schemas/bytebase.v1.AIChatToolCall'
title: tool_calls
description: Tool calls the AI wants to make.
+ usage:
+ oneOf:
+ - $ref: '#/components/schemas/bytebase.v1.AIChatUsage'
+ - type: "null"
+ title: usage
+ description: Token usage for this provider call, when available.
title: AIChatResponse
additionalProperties: false
description: Response message for AIService.Chat.
@@ -16258,6 +16264,17 @@ components:
title: AIChatToolDefinition
additionalProperties: false
description: A tool definition that the AI can invoke.
+ bytebase.v1.AIChatUsage:
+ type: object
+ properties:
+ totalTokens:
+ type: integer
+ title: total_tokens
+ format: int32
+ description: Total tokens used by the provider call.
+ title: AIChatUsage
+ additionalProperties: false
+ description: Token usage for a single AI provider call.
bytebase.v1.AICompletionRequest:
type: object
properties:
diff --git a/proto/gen/grpc-doc/v1/README.md b/proto/gen/grpc-doc/v1/README.md
index 8dac9b391d..cecbd82a11 100644
--- a/proto/gen/grpc-doc/v1/README.md
+++ b/proto/gen/grpc-doc/v1/README.md
@@ -140,6 +140,7 @@
- [AIChatResponse](#bytebase-v1-AIChatResponse)
- [AIChatToolCall](#bytebase-v1-AIChatToolCall)
- [AIChatToolDefinition](#bytebase-v1-AIChatToolDefinition)
+ - [AIChatUsage](#bytebase-v1-AIChatUsage)
- [AIChatMessageRole](#bytebase-v1-AIChatMessageRole)
@@ -2739,6 +2740,7 @@ Response message for AIService.Chat.
| ----- | ---- | ----- | ----------- |
| content | [string](#string) | optional | The text content of the AI response. Optional when the response only contains tool calls. |
| tool_calls | [AIChatToolCall](#bytebase-v1-AIChatToolCall) | repeated | Tool calls the AI wants to make. |
+| usage | [AIChatUsage](#bytebase-v1-AIChatUsage) | optional | Token usage for this provider call, when available. |
@@ -2779,6 +2781,21 @@ A tool definition that the AI can invoke.
+
+<a name="bytebase-v1-AIChatUsage"></a>
+
+### AIChatUsage
+Token usage for a single AI provider call.
+
+
+| Field | Type | Label | Description |
+| ----- | ---- | ----- | ----------- |
+| total_tokens | [int32](#int32) | | Total tokens used by the provider call. |
+
+
+
+
+
diff --git a/proto/gen/grpc-doc/v1/index.html b/proto/gen/grpc-doc/v1/index.html
index bf3d6da9f3..3acfe1f241 100644
--- a/proto/gen/grpc-doc/v1/index.html
+++ b/proto/gen/grpc-doc/v1/index.html
@@ -704,6 +704,10 @@
<a href="#bytebase.v1.AIChatToolDefinition"><span class="badge">M</span>AIChatToolDefinition</a>
</li>
+ <li>
+ <a href="#bytebase.v1.AIChatUsage"><span class="badge">M</span>AIChatUsage</a>
+ </li>
+
<li>
<a href="#bytebase.v1.AIChatMessageRole"><span class="badge">E</span>AIChatMessageRole</a>
@@ -8328,6 +8332,13 @@ Permissions required: None</p></td>
<td><p>Tool calls the AI wants to make. </p></td>
</tr>
+ <tr>
+ <td>usage</td>
+ <td><a href="#bytebase.v1.AIChatUsage">AIChatUsage</a></td>
+ <td>optional</td>
+ <td><p>Token usage for this provider call, when available. </p></td>
+ </tr>
+
</tbody>
</table>
@@ -8419,6 +8430,30 @@ Frontend must echo this back unchanged when sending tool results. </p></td>
+ <h3 id="bytebase.v1.AIChatUsage">AIChatUsage</h3>
+ <p>Token usage for a single AI provider call.</p>
+
+
+ <table class="field-table">
+ <thead>
+ <tr><td>Field</td><td>Type</td><td>Label</td><td>Description</td></tr>
+ </thead>
+ <tbody>
+
+ <tr>
+ <td>total_tokens</td>
+ <td><a href="#int32">int32</a></td>
+ <td></td>
+ <td><p>Total tokens used by the provider call. </p></td>
+ </tr>
+
+ </tbody>
+ </table>
+
+
+
+
+
<h3 id="bytebase.v1.AIChatMessageRole">AIChatMessageRole</h3>
diff --git a/proto/v1/v1/ai_service.proto b/proto/v1/v1/ai_service.proto
index c4c0a36e16..332c7bc47d 100644
--- a/proto/v1/v1/ai_service.proto
+++ b/proto/v1/v1/ai_service.proto
@@ -86,6 +86,12 @@ message AIChatRequest {
repeated AIChatToolDefinition tool_definitions = 2;
}
+// Token usage for a single AI provider call.
+message AIChatUsage {
+ // Total tokens used by the provider call.
+ int32 total_tokens = 1;
+}
+
// Response message for AIService.Chat.
message AIChatResponse {
// The text content of the AI response. Optional when the response only contains tool calls.
@@ -93,4 +99,7 @@ message AIChatResponse {
// Tool calls the AI wants to make.
repeated AIChatToolCall tool_calls = 2;
+
+ // Token usage for this provider call, when available.
+ optional AIChatUsage usage = 3;
}