1

2

3

export TYPESAFE_API_KEY="你的密钥"

pip install typesafe-sdk        # Python 3.10 以上

npm install @typesafe-ai/sdk    # Node.js 20 以上

1

2

3

4

5

6

7

8

9

10

11

12

13

curl -X POST https://api.typesafe.ai/v1/systemone \

  -H "Authorization: Bearer $TYPESAFE_API_KEY" \

  -H "Content-Type: application/json" \

  -d '{

    "state": "Hi, I have been trying to connect my Stripe account for 3 days and the integration keeps failing. I am losing sales. Please help ASAP.",

    "model": "jev-latest",

    "questions": {

      "urgency": {

        "type": "noul",

        "instructions": "Does this message express urgency?"

      }

    }

  }'

1

2

3

4

5

6

7

{

  "model": "jev-1.13.0",

  "answers": {

    "urgency": { "type": "noul", "noul": 0.99 }

  },

  "usage": { "input_tokens": 360, "output_tokens": 39 }

}

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

from typesafe_sdk import Choice, Noul, Score, TypeSafeClient

ticket = "Hi, I have been trying to connect my Stripe account for 3 days and the integration keeps failing. I am losing sales. Please help ASAP."

with TypeSafeClient() as client:

    response = client.system_one(

        state=ticket,

        questions={

            "department": Choice(

                instructions="Which team should handle this",

                criteria={

                    "billing": "Payments, invoices, refunds",

                    "technical": "Integration errors, API failures, bugs",

                    "sales": "Pricing questions, upgrades, new purchases",

                },

            ),

            "frustration": Score(

                instructions="How frustrated is the customer",

                criteria=[

                    "Neutral or polite, no complaint",

                    "Annoyed, mentions a problem but stays civil",

                    "Angry, threatens to leave or uses hostile language",

                ],

            ),

            "is_urgent": Noul(instructions="Does this message express urgency?"),

        },

    )

print(response.answers["department"].choice)       # technical

print(response.answers["frustration"].score)       # 1.0

print(response.answers["is_urgent"].noul)          # 1.0

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

from typesafe_sdk import Choice, TypeSafeClient

with TypeSafeClient() as client:

    response = client.system_one(

        state=user_message,

        questions={

            "action": Choice(

                instructions="What does the user want to do",

                criteria={

                    "check_balance": "View account balance",

                    "approve_transfer": "Approve a pending withdrawal",

                    "support": "Get help with something else",

                },

            )

        },

    )

# route_to_human、show_balance 等为业务侧自定义函数

answer = response.answers["action"]

if answer.confidence < 0.5:

    route_to_human(user_message)            # 模型确实不确定，不要猜

elif answer.choice == "check_balance":

    show_balance(account_id)                # 低风险、可恢复，直接执行

elif answer.choice == "approve_transfer":

    if answer.confidence > 0.9:

        confirm_then_execute(account_id)    # 高风险操作要求更高门槛

    else:

        ask_user_to_confirm(account_id)

1

pip install 'system-one-adapter[openai]'

1

2

3

4

5

6

7

8

9

10

11

12

13

14

from system_one_adapter import SystemOneAdapterClient, Noul

from system_one_adapter.providers.openai import OpenAIProvider

client = SystemOneAdapterClient(

    structured_outputs=True,

    llm_answer_mode="probabilities",

    normalize_probabilities=True,

)

response = client.system_one(

    state="This book was a delight to read.",

    questions={"positive": Noul(instructions="The book review is positive.")},

    model=OpenAIProvider("模型名", base_url="https://你的兼容端点/v1"),

)

print(response.answers["positive"].noul)

print(response.usage.latency, response.usage.input_tokens_total)