建立決策¶
此 API 為 Beta 版。端點位於 /v1beta/.
POST https://generativelanguage.googleapis.com/v1beta/decisions
針對共用輸入,評估一整批結構化決策問題。
每個問題都會獨立作答,回應會針對每個問題提供一個經過校準的答案,並以請求中的問題 ID 作為對應鍵。
請求主體¶
model: required string
用於評估決策的 Model 名稱。
**input**: required `DecisionInputContent` or `DecisionInputContent array` or `string` 所有決策問題共用的輸入前置內容,只評估一次。
**可能的類型**:**AudioContent**
音訊內容區塊。 **type**: required `object` 值:`audio` **data**: `string` 音訊內容。 **uri**: `string` 音訊的 URI。 **channels**: `integer` 音訊聲道數量。 **sample_rate**: `integer` 音訊的取樣率。**mime_type**: `string enum` 音訊的 MIME 類型。
可能的值: - `audio/wav`:WAV 音訊格式。 - `audio/mp3`:MP3 音訊格式。 - `audio/aiff`:AIFF 音訊格式。 - `audio/aac`:AAC 音訊格式。 - `audio/ogg`:OGG 音訊格式。 - `audio/flac`:FLAC 音訊格式。 - `audio/mpeg`:MPEG 音訊格式。 - `audio/m4a`:M4A 音訊格式。 - `audio/l16`:L16 音訊格式。 - `audio/opus`:OPUS 音訊格式。 - `audio/alaw`:ALAW 音訊格式。 - `audio/mulaw`:MULAW 音訊格式。 - `audio/webm`:WEBM 音訊格式。**DocumentContent**
文件內容區塊。 **type**: required `object` 值:`document` **data**: `string` 文件內容。 **uri**: `string` 文件的 URI。**mime_type**: `string enum` 文件的 MIME 類型。
可能的值: - `application/pdf`:PDF 文件格式。 - `text/csv`:CSV 文件格式。 - `application/vnd.openxmlformats-officedocument.presentationml.presentation`:PPTX 簡報格式。 - `application/vnd.ms-powerpoint`:PPT 簡報格式。 - `application/vnd.oasis.opendocument.presentation`:ODP 簡報格式。 - `application/vnd.openxmlformats-officedocument.wordprocessingml.document`:DOCX 文件格式。 - `application/msword`:DOC 文件格式。 - `application/vnd.openxmlformats-officedocument.wordprocessingml.template`:DOTX 文件範本格式。 - `application/vnd.oasis.opendocument.text`:ODT 文件格式。 - `application/vnd.openxmlformats-officedocument.spreadsheetml.sheet`:XLSX 試算表格式。 - `application/vnd.ms-excel`:XLS 試算表格式。 - `application/vnd.oasis.opendocument.spreadsheet`:ODS 試算表格式。**ImageContent**
圖片內容區塊。 **type**: required `object` 值:`image` **data**: `string` 圖片內容。 **uri**: `string` 圖片的 URI。**resolution**: `MediaResolution` 媒體的解析度。
可能的值: - `low`:低解析度。 - `medium`:中解析度。 - `high`:高解析度。 - `ultra_high`:超高解析度。**mime_type**: `string enum` 圖片的 MIME 類型。
可能的值: - `image/png`:PNG 圖片格式。 - `image/jpeg`:JPEG 圖片格式。 - `image/webp`:WebP 圖片格式。 - `image/heic`:HEIC 圖片格式。 - `image/heif`:HEIF 圖片格式。 - `image/gif`:GIF 圖片格式。 - `image/bmp`:BMP 圖片格式。 - `image/tiff`:TIFF 圖片格式。**TextContent**
文字內容區塊。 **text**: required `string` 文字內容。 **type**: required `object` 值:`text`**annotations**: `Annotation array` 模型生成內容的引用資訊。
**可能的類型**:**FileCitation**
檔案引用註記。 **type**: required `object` 值:`file_citation` **document_uri**: `string` 檔案的 URI。 **file_name**: `string` 檔案名稱。 **source**: `string` 這段文字所歸屬的來源。 **custom_metadata**: `object` 使用者提供的擷取上下文相關中繼資料。 **page_number**: `integer` 所引用文件的頁碼(如適用)。 **media_id**: `string` 圖片引用時的媒體 ID(如適用)。 **start_index**: `integer` 回應中歸屬此來源的區段起點。索引表示區段的開始位置,以位元組為單位。 **end_index**: `integer` 所歸屬區段的終點(不含該位置)。**PlaceCitation**
地點引用註記。 **type**: required `object` 值:`place_citation` **place_id**: `string` 地點的 ID,格式為 `places/{place_id}`。 **name**: `string` 地點的標題。 **url**: `string` 地點的 URI 參照。**review_snippets**: `ReviewSnippet array` 用來生成 Google 地圖中特定地點特色相關回答的評論片段。
**title**: `string` 評論的標題。 **url**: `string` 對應到 Google 地圖上使用者評論的連結。 **review_id**: `string` 評論片段的 ID。**SpeechAnnotation**
文字內容的語音註記。 **type**: required `object` 值:`speech_metadata` **speaker**: `string` 要與這個發言回合建立關聯的說話者。 **style**: `string` 語音合成的風格指令。 **start_index**: `integer` 回應中歸屬此來源的區段起點。索引表示區段的開始位置,以位元組為單位。 **end_index**: `integer` 所歸屬區段的終點(不含該位置)。**UrlCitation**
URL 引用註記。 **type**: required `object` 值:`url_citation` **url**: `string` URL。 **title**: `string` URL 的標題。 **start_index**: `integer` 回應中歸屬此來源的區段起點。索引表示區段的開始位置,以位元組為單位。 **end_index**: `integer` 所歸屬區段的終點(不含該位置)。**WordInfo**
用於轉錄輸出的單字層級 ASR 註記。包含單字文字、選填的時間資訊,以及選填的說話者歸屬。 **type**: required `object` 值:`word_info` **text**: `string` 轉錄出的單字。 **start_offset**: `string` 單字相對於音訊開頭的開始時間偏移。當 timestamp_granularities 包含 "word" 時才會出現。 **end_offset**: `string` 單字相對於音訊開頭的結束時間偏移。當 timestamp_granularities 包含 "word" 時才會出現。 **speaker**: `string` 選填。此單字的說話者標籤(例如 "spk_1"、"spk_2")。在 TranscriptionConfig 中設定了 diarization_mode 時才會出現。 **start_index**: `integer` 回應中歸屬此來源的區段起點。索引表示區段的開始位置,以位元組為單位。 **end_index**: `integer` 所歸屬區段的終點(不含該位置)。**VideoContent**
影片內容區塊。 **type**: required `object` 值:`video` **data**: `string` 影片內容。 **uri**: `string` 影片的 URI。**resolution**: `MediaResolution` 媒體的解析度。
可能的值: - `low`:低解析度。 - `medium`:中解析度。 - `high`:高解析度。 - `ultra_high`:超高解析度。**processing**: `MediaProcessing` or `string enum` 模型如何處理這部影片以進行理解。
**可能的類型**:**MediaProcessing**
**type**: required `object` 值:`static` **start_offset**: `string` 選填。區段開始時間。以十進位秒數表示,後面加上 's' 後綴,例如 "10.5s"。必須為非負數。 **end_offset**: `string` 選填。區段結束時間。以十進位秒數表示,後面加上 's' 後綴,例如 "30s"。必須為非負數,且在設定了 `start_offset` 時,必須大於 `start_offset`。 **fps**: `number` 選填。影片影格率取樣密度。**string enum**
可能的值: - `static`:以固定速率擷取影格。所有影格都會放入上下文中。 - `agentic`:由模型驅動的動態導覽。**mime_type**: `string enum` 影片的 MIME 類型。
可能的值: - `video/mp4`:MP4 影片格式。 - `video/mpeg`:MPEG 影片格式。 - `video/mpg`:MPG 影片格式。 - `video/mov`:MOV 影片格式。 - `video/avi`:AVI 影片格式。 - `video/x-flv`:FLV 影片格式。 - `video/webm`:WebM 影片格式。 - `video/wmv`:WMV 影片格式。 - `video/3gpp`:3GPP 影片格式。**DecisionInputContent array**
清單中的每個項目為下列其中之一:`AudioContent`、`DocumentContent`、`ImageContent`、`TextContent`、`VideoContent`。**string**
直接的字串值。questions: required object
要評估的決策問題,以呼叫端定義的唯一問題識別碼作為對應鍵(至少 1 個問題)。
範例請求
curl -X POST https://generativelanguage.googleapis.com/v1beta/decisions \
-H "x-goog-api-key: $GEMINI_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "gemini-decision-1",
"input": "Customer message: Hi, I was charged twice for my monthly subscription this morning. Can you refund the duplicate charge?",
"questions": {
"needs_billing_refund": {
"type": "noul",
"instructions": "Determine whether the customer is requesting a billing refund.",
"criteria": {
"true_description": "The customer explicitly asks for a refund or reversal of a charge.",
"false_description": "The customer is asking a general question or reporting a non-billing issue."
}
}
}
}'
from google import genai
client = genai.Client(http_options={"api_version": "v1beta"})
response = client.decisions.create(
model="gemini-decision-1",
input=(
"Customer message: Hi, I was charged twice for my monthly"
" subscription this morning. Can you refund the duplicate charge?"
),
questions={
"needs_billing_refund": {
"type": "noul",
"instructions": (
"Determine whether the customer is requesting a billing"
" refund."
),
"criteria": {
"true_description": (
"The customer explicitly asks for a refund or reversal"
" of a charge."
),
"false_description": (
"The customer is asking a general question or reporting"
" a non-billing issue."
),
},
},
},
)
if response.answers:
answer = response.answers["needs_billing_refund"]
if answer.type == "noul":
print(f"P(true)={answer.noul}, confidence={answer.confidence}")
import {GoogleGenAI} from '@google/genai';
const ai = new GoogleGenAI({apiVersion: 'v1beta'});
const response = await ai.decisions.create({
model: 'gemini-decision-1',
input:
'Customer message: Hi, I was charged twice for my monthly subscription this morning. Can you refund the duplicate charge?',
questions: {
needs_billing_refund: {
type: 'noul',
instructions:
'Determine whether the customer is requesting a billing refund.',
criteria: {
true_description:
'The customer explicitly asks for a refund or reversal of a charge.',
false_description:
'The customer is asking a general question or reporting a non-billing issue.',
},
},
},
});
const answer = response.answers?.['needs_billing_refund'];
if (answer?.type === 'noul') {
console.log(`P(true)=${answer.noul}, confidence=${answer.confidence}`);
}
import com.google.genai.gaos.models.decisions.Answer;
import com.google.genai.Client;
import com.google.genai.gaos.models.decisions.CreateDecisionRequest;
import com.google.genai.gaos.models.operations.CreateDecisionResponse;
import com.google.genai.types.HttpOptions;
import com.google.genai.gaos.models.decisions.Input;
import com.google.genai.gaos.models.decisions.NoulAnswer;
import com.google.genai.gaos.models.decisions.NoulCriteria;
import com.google.genai.gaos.models.decisions.NoulQuestion;
import com.google.genai.gaos.models.decisions.NoulQuestionInstructions;
Client client =
Client.builder()
.httpOptions(HttpOptions.builder().apiVersion("v1beta").build())
.build();
CreateDecisionRequest request =
CreateDecisionRequest.builder()
.model("gemini-decision-1")
.input(
Input.of(
"Customer message: Hi, I was charged twice for my monthly subscription this"
+ " morning. Can you refund the duplicate charge?"))
.questions(
Map.of(
"needs_billing_refund",
NoulQuestion.builder()
.instructions(
NoulQuestionInstructions.of(
"Determine whether the customer is requesting a billing refund."))
.criteria(
NoulCriteria.builder()
.trueDescription(
"The customer explicitly asks for a refund or reversal of a"
+ " charge.")
.falseDescription(
"The customer is asking a general question or reporting a"
+ " non-billing issue.")
.build())
.build()))
.build();
CreateDecisionResponse response = client.decisions.create(request);
Map<String, Answer> answers =
response.createDecisionResponse().flatMap(r -> r.answers()).orElse(Map.of());
Answer answer = answers.get("needs_billing_refund");
if (answer instanceof NoulAnswer noulAnswer) {
System.out.println(
"P(true)="
+ noulAnswer.noul().orElse(0.0)
+ ", confidence="
+ noulAnswer.confidence().orElse(0.0));
}
package main
import (
"context"
"fmt"
"log"
"google.golang.org/genai"
gaos_decisions "google.golang.org/genai/interactions/models/decisions"
"google.golang.org/genai/interactions/models/operations"
)
func main() {
ctx := context.Background()
client, err := genai.NewClient(ctx, &genai.ClientConfig{
HTTPOptions: genai.HTTPOptions{APIVersion: "v1beta"},
})
if err != nil {
log.Fatal(err)
}
req := operations.CreateDecisionRequest{
Body: gaos_decisions.CreateDecisionRequest{
Model: "gemini-decision-1",
Input: gaos_decisions.NewInput("Customer message: Hi, I was charged twice for my monthly subscription this morning. Can you refund the duplicate charge?"),
Questions: map[string]gaos_decisions.Question{
"needs_billing_refund": gaos_decisions.NewQuestion(gaos_decisions.NoulQuestion{
Instructions: gaos_decisions.NewNoulQuestionInstructions("Determine whether the customer is requesting a billing refund."),
Criteria: &gaos_decisions.NoulCriteria{
TrueDescription: genai.Ptr("The customer explicitly asks for a refund or reversal of a charge."),
FalseDescription: genai.Ptr("The customer is asking a general question or reporting a non-billing issue."),
},
}),
},
},
}
res, err := client.Decisions.Create(ctx, req)
if err != nil {
log.Fatal(err)
}
if res.CreateDecisionResponse != nil {
if answer, ok := res.CreateDecisionResponse.Answers["needs_billing_refund"]; ok && answer.NoulAnswer != nil {
fmt.Printf("P(true)=%v, confidence=%v\n", answer.NoulAnswer.GetNoul(), answer.NoulAnswer.GetConfidence())
}
}
}
curl -X POST https://generativelanguage.googleapis.com/v1beta/decisions \
-H "x-goog-api-key: $GEMINI_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "gemini-decision-1",
"input": "Customer message: Every time I click Export to PDF, the app freezes and crashes with error code 504.",
"questions": {
"ticket_route": {
"type": "choice",
"instructions": "Route the support ticket to the most appropriate team.",
"criteria": {
"billing": "Invoices, charges, refunds, or subscription plan changes.",
"technical_support": "Bugs, crashes, errors, or product functionality issues.",
"account_access": "Login, password reset, or two-factor authentication issues."
}
}
}
}'
from google import genai
client = genai.Client(http_options={"api_version": "v1beta"})
response = client.decisions.create(
model="gemini-decision-1",
input=(
"Customer message: Every time I click Export to PDF, the app"
" freezes and crashes with error code 504."
),
questions={
"ticket_route": {
"type": "choice",
"instructions": (
"Route the support ticket to the most appropriate team."
),
"criteria": {
"billing": (
"Invoices, charges, refunds, or subscription plan"
" changes."
),
"technical_support": (
"Bugs, crashes, errors, or product functionality"
" issues."
),
"account_access": (
"Login, password reset, or two-factor authentication"
" issues."
),
},
},
},
)
if response.answers:
answer = response.answers["ticket_route"]
if answer.type == "choice":
print(
f"choice={answer.choice}, confidence={answer.confidence},"
f" probabilities={answer.probabilities}"
)
import {GoogleGenAI} from '@google/genai';
const ai = new GoogleGenAI({apiVersion: 'v1beta'});
const response = await ai.decisions.create({
model: 'gemini-decision-1',
input:
'Customer message: Every time I click Export to PDF, the app freezes and crashes with error code 504.',
questions: {
ticket_route: {
type: 'choice',
instructions:
'Route the support ticket to the most appropriate team.',
criteria: {
billing:
'Invoices, charges, refunds, or subscription plan changes.',
technical_support:
'Bugs, crashes, errors, or product functionality issues.',
account_access:
'Login, password reset, or two-factor authentication issues.',
},
},
},
});
const answer = response.answers?.['ticket_route'];
if (answer?.type === 'choice') {
console.log(
`choice=${answer.choice}, confidence=${answer.confidence}`,
answer.probabilities,
);
}
import com.google.genai.gaos.models.decisions.Answer;
import com.google.genai.gaos.models.decisions.ChoiceAnswer;
import com.google.genai.gaos.models.decisions.ChoiceQuestion;
import com.google.genai.gaos.models.decisions.ChoiceQuestionInstructions;
import com.google.genai.Client;
import com.google.genai.gaos.models.decisions.CreateDecisionRequest;
import com.google.genai.gaos.models.operations.CreateDecisionResponse;
import com.google.genai.types.HttpOptions;
import com.google.genai.gaos.models.decisions.Input;
Client client =
Client.builder()
.httpOptions(HttpOptions.builder().apiVersion("v1beta").build())
.build();
CreateDecisionRequest request =
CreateDecisionRequest.builder()
.model("gemini-decision-1")
.input(
Input.of(
"Customer message: Every time I click Export to PDF, the app freezes and"
+ " crashes with error code 504."))
.questions(
Map.of(
"ticket_route",
ChoiceQuestion.builder()
.instructions(
ChoiceQuestionInstructions.of(
"Route the support ticket to the most appropriate team."))
.criteria(
Map.of(
"billing",
"Invoices, charges, refunds, or subscription plan changes.",
"technical_support",
"Bugs, crashes, errors, or product functionality issues.",
"account_access",
"Login, password reset, or two-factor authentication issues."))
.build()))
.build();
CreateDecisionResponse response = client.decisions.create(request);
Map<String, Answer> answers =
response.createDecisionResponse().flatMap(r -> r.answers()).orElse(Map.of());
Answer answer = answers.get("ticket_route");
if (answer instanceof ChoiceAnswer choiceAnswer) {
System.out.println(
"choice="
+ choiceAnswer.choice().orElse("")
+ ", confidence="
+ choiceAnswer.confidence().orElse(0.0)
+ ", probabilities="
+ choiceAnswer.probabilities().orElse(Map.of()));
}
package main
import (
"context"
"fmt"
"log"
"google.golang.org/genai"
gaos_decisions "google.golang.org/genai/interactions/models/decisions"
"google.golang.org/genai/interactions/models/operations"
)
func main() {
ctx := context.Background()
client, err := genai.NewClient(ctx, &genai.ClientConfig{
HTTPOptions: genai.HTTPOptions{APIVersion: "v1beta"},
})
if err != nil {
log.Fatal(err)
}
req := operations.CreateDecisionRequest{
Body: gaos_decisions.CreateDecisionRequest{
Model: "gemini-decision-1",
Input: gaos_decisions.NewInput("Customer message: Every time I click Export to PDF, the app freezes and crashes with error code 504."),
Questions: map[string]gaos_decisions.Question{
"ticket_route": gaos_decisions.NewQuestion(gaos_decisions.ChoiceQuestion{
Instructions: gaos_decisions.NewChoiceQuestionInstructions("Route the support ticket to the most appropriate team."),
Criteria: map[string]string{
"billing": "Invoices, charges, refunds, or subscription plan changes.",
"technical_support": "Bugs, crashes, errors, or product functionality issues.",
"account_access": "Login, password reset, or two-factor authentication issues.",
},
}),
},
},
}
res, err := client.Decisions.Create(ctx, req)
if err != nil {
log.Fatal(err)
}
if res.CreateDecisionResponse != nil {
if answer, ok := res.CreateDecisionResponse.Answers["ticket_route"]; ok && answer.ChoiceAnswer != nil {
fmt.Printf("choice=%s, confidence=%v, probabilities=%v\n", answer.ChoiceAnswer.GetChoice(), answer.ChoiceAnswer.GetConfidence(), answer.ChoiceAnswer.GetProbabilities())
}
}
}
curl -X POST https://generativelanguage.googleapis.com/v1beta/decisions \
-H "x-goog-api-key: $GEMINI_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "gemini-decision-1",
"input": "Draft reply: Duplicate charge refunded ($19.99) to your Visa ending in 4242. It will appear on your statement within 3-5 business days.",
"questions": {
"clarity": {
"type": "score",
"instructions": "Score the clarity and actionability of the customer support reply.",
"criteria": [
"Unclear, confusing, or missing key details.",
"Partially clear, but leaves important questions unanswered.",
"Clear, concise, and gives concrete details and timeline."
]
}
}
}'
from google import genai
client = genai.Client(http_options={"api_version": "v1beta"})
response = client.decisions.create(
model="gemini-decision-1",
input=(
"Draft reply: Duplicate charge refunded ($19.99) to your Visa"
" ending in 4242. It will appear on your statement within 3-5"
" business days."
),
questions={
"clarity": {
"type": "score",
"instructions": (
"Score the clarity and actionability of the customer"
" support reply."
),
"criteria": [
"Unclear, confusing, or missing key details.",
"Partially clear, but leaves important questions unanswered.",
"Clear, concise, and gives concrete details and timeline.",
],
},
},
)
if response.answers:
answer = response.answers["clarity"]
if answer.type == "score":
print(
f"score={answer.score}, confidence={answer.confidence},"
f" probabilities={answer.probabilities}"
)
import {GoogleGenAI} from '@google/genai';
const ai = new GoogleGenAI({apiVersion: 'v1beta'});
const response = await ai.decisions.create({
model: 'gemini-decision-1',
input:
'Draft reply: Duplicate charge refunded ($19.99) to your Visa ending in 4242. It will appear on your statement within 3-5 business days.',
questions: {
clarity: {
type: 'score',
instructions:
'Score the clarity and actionability of the customer support reply.',
criteria: [
'Unclear, confusing, or missing key details.',
'Partially clear, but leaves important questions unanswered.',
'Clear, concise, and gives concrete details and timeline.',
],
},
},
});
const answer = response.answers?.['clarity'];
if (answer?.type === 'score') {
console.log(
`score=${answer.score}, confidence=${answer.confidence}`,
answer.probabilities,
);
}
import com.google.genai.gaos.models.decisions.Answer;
import com.google.genai.Client;
import com.google.genai.gaos.models.decisions.CreateDecisionRequest;
import com.google.genai.gaos.models.operations.CreateDecisionResponse;
import com.google.genai.types.HttpOptions;
import com.google.genai.gaos.models.decisions.Input;
import com.google.genai.gaos.models.decisions.ScoreAnswer;
import com.google.genai.gaos.models.decisions.ScoreQuestion;
import com.google.genai.gaos.models.decisions.ScoreQuestionInstructions;
Client client =
Client.builder()
.httpOptions(HttpOptions.builder().apiVersion("v1beta").build())
.build();
CreateDecisionRequest request =
CreateDecisionRequest.builder()
.model("gemini-decision-1")
.input(
Input.of(
"Draft reply: Duplicate charge refunded ($19.99) to your Visa ending in 4242."
+ " It will appear on your statement within 3-5 business days."))
.questions(
Map.of(
"clarity",
ScoreQuestion.builder()
.instructions(
ScoreQuestionInstructions.of(
"Score the clarity and actionability of the customer support"
+ " reply."))
.criteria(
List.of(
"Unclear, confusing, or missing key details.",
"Partially clear, but leaves important questions unanswered.",
"Clear, concise, and gives concrete details and timeline."))
.build()))
.build();
CreateDecisionResponse response = client.decisions.create(request);
Map<String, Answer> answers =
response.createDecisionResponse().flatMap(r -> r.answers()).orElse(Map.of());
Answer answer = answers.get("clarity");
if (answer instanceof ScoreAnswer scoreAnswer) {
System.out.println(
"score="
+ scoreAnswer.score().orElse(0.0)
+ ", confidence="
+ scoreAnswer.confidence().orElse(0.0)
+ ", probabilities="
+ scoreAnswer.probabilities().orElse(Map.of()));
}
package main
import (
"context"
"fmt"
"log"
"google.golang.org/genai"
gaos_decisions "google.golang.org/genai/interactions/models/decisions"
"google.golang.org/genai/interactions/models/operations"
)
func main() {
ctx := context.Background()
client, err := genai.NewClient(ctx, &genai.ClientConfig{
HTTPOptions: genai.HTTPOptions{APIVersion: "v1beta"},
})
if err != nil {
log.Fatal(err)
}
req := operations.CreateDecisionRequest{
Body: gaos_decisions.CreateDecisionRequest{
Model: "gemini-decision-1",
Input: gaos_decisions.NewInput("Draft reply: Duplicate charge refunded ($19.99) to your Visa ending in 4242. It will appear on your statement within 3-5 business days."),
Questions: map[string]gaos_decisions.Question{
"clarity": gaos_decisions.NewQuestion(gaos_decisions.ScoreQuestion{
Instructions: gaos_decisions.NewScoreQuestionInstructions("Score the clarity and actionability of the customer support reply."),
Criteria: []string{
"Unclear, confusing, or missing key details.",
"Partially clear, but leaves important questions unanswered.",
"Clear, concise, and gives concrete details and timeline.",
},
}),
},
},
}
res, err := client.Decisions.Create(ctx, req)
if err != nil {
log.Fatal(err)
}
if res.CreateDecisionResponse != nil {
if answer, ok := res.CreateDecisionResponse.Answers["clarity"]; ok && answer.ScoreAnswer != nil {
fmt.Printf("score=%v, confidence=%v, probabilities=%v\n", answer.ScoreAnswer.GetScore(), answer.ScoreAnswer.GetConfidence(), answer.ScoreAnswer.GetProbabilities())
}
}
}
curl -X POST https://generativelanguage.googleapis.com/v1beta/decisions \
-H "x-goog-api-key: $GEMINI_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "gemini-decision-1",
"input": "Customer message: Production API requests have been failing with 503 errors for the past 20 minutes, blocking all checkout traffic!",
"questions": {
"is_outage": {
"type": "noul",
"instructions": "Determine whether the customer is reporting an active production outage."
},
"priority": {
"type": "choice",
"instructions": "Select the incident priority level.",
"criteria": {
"p1_critical": "Production outage blocking core business traffic.",
"p2_high": "Major feature degraded with a workaround available.",
"p3_normal": "Minor bug or general question."
}
},
"urgency": {
"type": "score",
"instructions": "Rate the urgency of immediate escalation.",
"criteria": [
"Routine inquiry, standard SLA.",
"Time-sensitive issue needing same-day follow-up.",
"Immediate escalation required for active production impact."
]
}
}
}'
from google import genai
client = genai.Client(http_options={"api_version": "v1beta"})
response = client.decisions.create(
model="gemini-decision-1",
input=(
"Customer message: Production API requests have been failing with"
" 503 errors for the past 20 minutes, blocking all checkout"
" traffic!"
),
questions={
"is_outage": {
"type": "noul",
"instructions": (
"Determine whether the customer is reporting an active"
" production outage."
),
},
"priority": {
"type": "choice",
"instructions": "Select the incident priority level.",
"criteria": {
"p1_critical": (
"Production outage blocking core business traffic."
),
"p2_high": (
"Major feature degraded with a workaround available."
),
"p3_normal": "Minor bug or general question.",
},
},
"urgency": {
"type": "score",
"instructions": "Rate the urgency of immediate escalation.",
"criteria": [
"Routine inquiry, standard SLA.",
"Time-sensitive issue needing same-day follow-up.",
"Immediate escalation required for active production impact.",
],
},
},
)
for question_id, answer in (response.answers or {}).items():
if answer.type == "noul":
print(f"{question_id}: P(true)={answer.noul}")
elif answer.type == "choice":
print(f"{question_id}: choice={answer.choice}")
elif answer.type == "score":
print(f"{question_id}: score={answer.score}")
import {GoogleGenAI} from '@google/genai';
const ai = new GoogleGenAI({apiVersion: 'v1beta'});
const response = await ai.decisions.create({
model: 'gemini-decision-1',
input:
'Customer message: Production API requests have been failing with 503 errors for the past 20 minutes, blocking all checkout traffic!',
questions: {
is_outage: {
type: 'noul',
instructions:
'Determine whether the customer is reporting an active production outage.',
},
priority: {
type: 'choice',
instructions: 'Select the incident priority level.',
criteria: {
p1_critical: 'Production outage blocking core business traffic.',
p2_high: 'Major feature degraded with a workaround available.',
p3_normal: 'Minor bug or general question.',
},
},
urgency: {
type: 'score',
instructions: 'Rate the urgency of immediate escalation.',
criteria: [
'Routine inquiry, standard SLA.',
'Time-sensitive issue needing same-day follow-up.',
'Immediate escalation required for active production impact.',
],
},
},
});
for (const [questionId, answer] of Object.entries(response.answers ?? {})) {
if (answer.type === 'noul') {
console.log(`${questionId}: P(true)=${answer.noul}`);
} else if (answer.type === 'choice') {
console.log(`${questionId}: choice=${answer.choice}`);
} else if (answer.type === 'score') {
console.log(`${questionId}: score=${answer.score}`);
}
}
import com.google.genai.gaos.models.decisions.Answer;
import com.google.genai.gaos.models.decisions.ChoiceAnswer;
import com.google.genai.gaos.models.decisions.ChoiceQuestion;
import com.google.genai.gaos.models.decisions.ChoiceQuestionInstructions;
import com.google.genai.Client;
import com.google.genai.gaos.models.decisions.CreateDecisionRequest;
import com.google.genai.gaos.models.operations.CreateDecisionResponse;
import com.google.genai.types.HttpOptions;
import com.google.genai.gaos.models.decisions.Input;
import com.google.genai.gaos.models.decisions.NoulAnswer;
import com.google.genai.gaos.models.decisions.NoulQuestion;
import com.google.genai.gaos.models.decisions.NoulQuestionInstructions;
import com.google.genai.gaos.models.decisions.ScoreAnswer;
import com.google.genai.gaos.models.decisions.ScoreQuestion;
import com.google.genai.gaos.models.decisions.ScoreQuestionInstructions;
Client client =
Client.builder()
.httpOptions(HttpOptions.builder().apiVersion("v1beta").build())
.build();
CreateDecisionRequest request =
CreateDecisionRequest.builder()
.model("gemini-decision-1")
.input(
Input.of(
"Customer message: Production API requests have been failing with 503 errors"
+ " for the past 20 minutes, blocking all checkout traffic!"))
.questions(
Map.of(
"is_outage",
NoulQuestion.builder()
.instructions(
NoulQuestionInstructions.of(
"Determine whether the customer is reporting an active production"
+ " outage."))
.build(),
"priority",
ChoiceQuestion.builder()
.instructions(
ChoiceQuestionInstructions.of("Select the incident priority level."))
.criteria(
Map.of(
"p1_critical",
"Production outage blocking core business traffic.",
"p2_high",
"Major feature degraded with a workaround available.",
"p3_normal",
"Minor bug or general question."))
.build(),
"urgency",
ScoreQuestion.builder()
.instructions(
ScoreQuestionInstructions.of(
"Rate the urgency of immediate escalation."))
.criteria(
List.of(
"Routine inquiry, standard SLA.",
"Time-sensitive issue needing same-day follow-up.",
"Immediate escalation required for active production impact."))
.build()))
.build();
CreateDecisionResponse response = client.decisions.create(request);
Map<String, Answer> answers =
response.createDecisionResponse().flatMap(r -> r.answers()).orElse(Map.of());
for (Map.Entry<String, Answer> entry : answers.entrySet()) {
Answer answer = entry.getValue();
if (answer instanceof NoulAnswer noulAnswer) {
System.out.println(entry.getKey() + ": P(true)=" + noulAnswer.noul().orElse(0.0));
} else if (answer instanceof ChoiceAnswer choiceAnswer) {
System.out.println(entry.getKey() + ": choice=" + choiceAnswer.choice().orElse(""));
} else if (answer instanceof ScoreAnswer scoreAnswer) {
System.out.println(entry.getKey() + ": score=" + scoreAnswer.score().orElse(0.0));
}
}
package main
import (
"context"
"fmt"
"log"
"google.golang.org/genai"
gaos_decisions "google.golang.org/genai/interactions/models/decisions"
"google.golang.org/genai/interactions/models/operations"
)
func main() {
ctx := context.Background()
client, err := genai.NewClient(ctx, &genai.ClientConfig{
HTTPOptions: genai.HTTPOptions{APIVersion: "v1beta"},
})
if err != nil {
log.Fatal(err)
}
req := operations.CreateDecisionRequest{
Body: gaos_decisions.CreateDecisionRequest{
Model: "gemini-decision-1",
Input: gaos_decisions.NewInput("Customer message: Production API requests have been failing with 503 errors for the past 20 minutes, blocking all checkout traffic!"),
Questions: map[string]gaos_decisions.Question{
"is_outage": gaos_decisions.NewQuestion(gaos_decisions.NoulQuestion{
Instructions: gaos_decisions.NewNoulQuestionInstructions("Determine whether the customer is reporting an active production outage."),
}),
"priority": gaos_decisions.NewQuestion(gaos_decisions.ChoiceQuestion{
Instructions: gaos_decisions.NewChoiceQuestionInstructions("Select the incident priority level."),
Criteria: map[string]string{
"p1_critical": "Production outage blocking core business traffic.",
"p2_high": "Major feature degraded with a workaround available.",
"p3_normal": "Minor bug or general question.",
},
}),
"urgency": gaos_decisions.NewQuestion(gaos_decisions.ScoreQuestion{
Instructions: gaos_decisions.NewScoreQuestionInstructions("Rate the urgency of immediate escalation."),
Criteria: []string{
"Routine inquiry, standard SLA.",
"Time-sensitive issue needing same-day follow-up.",
"Immediate escalation required for active production impact.",
},
}),
},
},
}
res, err := client.Decisions.Create(ctx, req)
if err != nil {
log.Fatal(err)
}
if res.CreateDecisionResponse != nil {
for questionID, answer := range res.CreateDecisionResponse.Answers {
if answer.NoulAnswer != nil {
fmt.Printf("%s: P(true)=%v\n", questionID, answer.NoulAnswer.GetNoul())
} else if answer.ChoiceAnswer != nil {
fmt.Printf("%s: choice=%s\n", questionID, answer.ChoiceAnswer.GetChoice())
} else if answer.ScoreAnswer != nil {
fmt.Printf("%s: score=%v\n", questionID, answer.ScoreAnswer.GetScore())
}
}
}
}
回應¶
answers: object
僅供輸出。決策答案,以 questions 中的問題識別碼作為對應鍵。
**usage**: `Usage` 僅供輸出。此決策評估請求的權杖用量統計。
**total_input_tokens**: `integer` 提示詞(上下文)中的權杖數量。 **initial_input_tokens**: `integer` 此互動第一個回合所使用的輸入權杖數量。**input_tokens_by_modality**: `ModalityTokens array` 依模態區分的輸入權杖用量明細。
**modality**: `ResponseModality` 與權杖數量對應的模態。
可能的值: - `text`:表示模型應傳回文字。 - `image`:表示模型應傳回圖片。 - `audio`:表示模型應傳回音訊。 - `video`:表示模型應傳回影片。 - `document`:表示模型應傳回文件。**cached_tokens_by_modality**: `ModalityTokens array` 依模態區分的快取權杖用量明細。
**modality**: `ResponseModality` 與權杖數量對應的模態。
可能的值: - `text`:表示模型應傳回文字。 - `image`:表示模型應傳回圖片。 - `audio`:表示模型應傳回音訊。 - `video`:表示模型應傳回影片。 - `document`:表示模型應傳回文件。**output_tokens_by_modality**: `ModalityTokens array` 依模態區分的輸出權杖用量明細。
**modality**: `ResponseModality` 與權杖數量對應的模態。
可能的值: - `text`:表示模型應傳回文字。 - `image`:表示模型應傳回圖片。 - `audio`:表示模型應傳回音訊。 - `video`:表示模型應傳回影片。 - `document`:表示模型應傳回文件。**tool_use_tokens_by_modality**: `ModalityTokens array` 依模態區分的工具使用權杖用量明細。
**modality**: `ResponseModality` 與權杖數量對應的模態。
可能的值: - `text`:表示模型應傳回文字。 - `image`:表示模型應傳回圖片。 - `audio`:表示模型應傳回音訊。 - `video`:表示模型應傳回影片。 - `document`:表示模型應傳回文件。**grounding_tool_count**: `GroundingToolCount array` 接地工具計數。
**type**: `string enum` 與計數對應的接地工具類型。
可能的值: - `google_search`:使用 Google 網頁搜尋與圖片搜尋接地,以及 Web Grounding for Enterprise。 - `google_maps`:使用 Google 地圖接地。prompt_feedback: PromptFeedback
僅供輸出。來自輸入內容篩選器的回饋。若設定了 prompt_feedback.block_reason,表示提示詞(input 與問題文字)已被輸入政策封鎖,不會評估任何決策,且 answers 為空。
範例回應
{
"answers": {
"needs_billing_refund": {
"type": "noul",
"confidence": 0.97,
"noul": 0.97
}
},
"usage": {
"input_tokens_by_modality": [
{
"modality": "text",
"tokens": 152
}
],
"total_input_tokens": 152,
"total_output_tokens": 0,
"total_tokens": 152
}
}
{
"answers": {
"ticket_route": {
"type": "choice",
"choice": "technical_support",
"confidence": 0.95,
"probabilities": {
"account_access": 0.02,
"billing": 0.03,
"technical_support": 0.95
}
}
}
}
{
"answers": {
"clarity": {
"type": "score",
"confidence": 0.82,
"legend": {
"0": "Unclear, confusing, or missing key details.",
"1": "Partially clear, but leaves important questions unanswered.",
"2": "Clear, concise, and gives concrete details and timeline."
},
"probabilities": {
"0": 0.02,
"1": 0.16,
"2": 0.82
},
"score": 1.8
}
}
}
{
"answers": {
"is_outage": {
"type": "noul",
"confidence": 0.98,
"noul": 0.98
},
"priority": {
"type": "choice",
"choice": "p1_critical",
"confidence": 0.93,
"probabilities": {
"p1_critical": 0.93,
"p2_high": 0.06,
"p3_normal": 0.01
}
},
"urgency": {
"type": "score",
"confidence": 0.91,
"legend": {
"0": "Routine inquiry, standard SLA.",
"1": "Time-sensitive issue needing same-day follow-up.",
"2": "Immediate escalation required for active production impact."
},
"probabilities": {
"0": 0.01,
"1": 0.08,
"2": 0.91
},
"score": 1.9
}
}
}
資料模型¶
PromptFeedback¶
來自輸入內容篩選器的回饋。
屬性:
**block_reason**: `string enum` 選填。若有設定,表示輸入已被封鎖,不會傳回任何答案。
可能的值: - `safety`:提示詞因安全因素遭到封鎖。 - `other`:提示詞因不明原因遭到封鎖。 - `blocklist`:提示詞因含有術語封鎖清單中的詞彙遭到封鎖。 - `prohibited_content`:提示詞因含有禁止內容遭到封鎖。 - `image_safety`:提示詞因圖片內容不安全遭到封鎖。block_reason_message: string
選填。詳細說明封鎖原因的訊息(選填)。