哑巴模型Jev使用完整指南：运行机制、原语定义、运行时环境配置及业务场景

1

2

python --version

node -v

1

export TYPESAFE_API_KEY="sk-your-api-key"

1

pip install typesafe-sdk

1

2

3

4

5

6

7

8

9

10

11

12

13

14

15

16

17

18

19

20

21

22

23

24

25

26

27

28

29

30

31

32

33

34

35

36

37

38

39

40

41

42

43

import os

from typesafe_sdk import TypeSafeClient, Choice, Score, Noul

# 客户端默认从环境变量 TYPESAFE_API_KEY 读取密钥

client = TypeSafeClient(api_key=os.environ.get("TYPESAFE_API_KEY"))

# 状态可以是字典、纯文本或多轮对话结构

state_data = {

    "ticket_id": "TCK-9801",

    "customer_message": "我已经连续三天无法导出账单了，系统一直报 500 错误，这已经影响到我们的对账审计，请立即处理。",

    "account_tier": "Enterprise"

}

# 组合并发提问的问题集合

questions_to_ask = {

    "target_team": Choice(

        instructions="确定负责处理此工单的业务团队",

        criteria={

            "billing": "涉及发票生成、支付扣款、套餐变更等账务问题",

            "technical": "平台功能故障、报错异常、API 接入问题",

            "sales": "商业报价与续约谈判",

            "other": "其余无法归类的情况"

        }

    ),

    "urgency_level": Score(

        instructions="评估用户反映问题的紧迫程度与情绪烈度",

        criteria=[

            "语气平缓，仅陈述事实",

            "表现出一定焦虑或受阻，但表达克制",

            "极其愤怒，包含业务受损警告或强硬措辞"

        ]

    ),

    "is_enterprise_blocker": Noul(

        instructions="客户是否在表达其业务流程受到阻断性影响"

    )

}

response = client.system_one(

    state=state_data,

    questions=questions_to_ask

)

team_result = response.answers["target_team"]

urgency_result = response.answers["urgency_level"]

blocker_result = response.answers["is_enterprise_blocker"]

print(f"分配部门: {team_result.choice} (置信度: {team_result.confidence:.2f})")

print(f"紧迫度评分: {urgency_result.score:.3f}")

print(f"阻断性概率: {blocker_result.noul:.2f}")

1

npm install @typesafe-ai/sdk

1

2

3

4

5

6

7

8

9

10

11

12

13

14

15

16

17

18

19

20

21

22

23

24

25

26

import { TypeSafeClient, choice, score, noul } from '@typesafe-ai/sdk';

const client = new TypeSafeClient();

const response = await client.systemOne({

  state: {

    ticket_id: "TCK-9801",

    customer_message: "我已经连续三天无法导出账单了，系统一直报 500 错误，这已经影响到我们的对账审计，请立即处理。",

    account_tier: "Enterprise"

  },

  questions: {

    target_team: choice("确定负责处理此工单的业务团队", {

      billing: "涉及发票生成、支付扣款、套餐变更等账务问题",

      technical: "平台功能故障、报错异常、API 接入问题",

      sales: "商业报价与续约谈判",

      other: "其余无法归类的情况"

    }),

    urgency_level: score("评估用户反映问题的紧迫程度与情绪烈度", [

      "语气平缓，仅陈述事实",

      "表现出一定焦虑或受阻，但表达克制",

      "极其愤怒，包含业务受损警告或强硬措辞"

    ]),

    is_enterprise_blocker: noul("客户是否在表达其业务流程受到阻断性影响")

  }

});

console.log(`分配部门: ${response.answers.target_team.choice}`);

console.log(`紧迫度评分: ${response.answers.urgency_level.score}`);

console.log(`阻断性概率: ${response.answers.is_enterprise_blocker.noul}`);

1

2

3

4

5

6

7

8

9

10

11

12

13

14

15

16

17

18

19

20

21

22

23

24

25

26

27

28

29

response = client.system_one(

    state=ticket_payload,

    questions={

        "category": Choice(

            instructions="工单所属主类别",

            criteria={

                "bug": "系统 Bug 报告",

                "refund": "退款请求",

                "consulting": "业务咨询"

            }

        ),

        "bug_reproducible": Noul(instructions="内容中是否包含可复现的步骤或环境说明"),

        "refund_eligible": Noul(instructions="客户是否明确要求退还已付金额"),

        "severity": Score(

            instructions="故障破坏力等级",

            criteria=["轻微UI缺陷", "部分功能受限有替代方案", "系统崩溃无替代方案"]

        )

    }

)

answers = response.answers

category = answers["category"].choice

# 由本地常规代码直接分发，无二次网络等待

if category == "bug":

    if answers["severity"].score > 1.2 and answers["bug_reproducible"].noul > 0.6:

        escalate_to_oncall_engineer(ticket_payload)

    else:

        append_to_backlog(ticket_payload)

elif category == "refund" and answers["refund_eligible"].noul > 0.8:

    trigger_refund_workflow(ticket_payload)

1

2

3

4

5

6

7

8

9

10

11

intent = response.answers["user_intent"]

# 置信度偏低时拒绝自动化盲目执行

if intent.confidence < 0.60:

    forward_to_human_agent(user_query)

elif intent.choice == "query_balance":

    execute_query_balance(user_id)

elif intent.choice == "transfer_funds":

    if intent.confidence >= 0.92:

        execute_transfer(transaction_data)

    else:

        prompt_user_secondary_confirmation(user_query)