API details

On May 3, 2024, Alibaba Cloud Model Studio was upgraded. For more information about the upgrade, see Official pre-built application upgrade announcement. After this upgrade, you can no longer create legacy RAG applications. If your application was created before this upgrade, you can continue to use this topic to call your application.

To learn how to migrate your enterprise knowledge base to the new knowledge base and build a new RAG application, see Migration plan below.
To learn how to call a new RAG application from your business code using an HTTP interface or a software development kit (SDK), see Call an application.

Migration plan

The enterprise knowledge base feature is no longer maintained. We recommend that you migrate your private knowledge to the new knowledge base as soon as possible. Alibaba Cloud does not currently provide official migration tools or services. The new knowledge base offers more features and supports more document formats.

Enterprise knowledge type	Migration plan
Documents	Create an unstructured knowledge base to manage document-based knowledge. Import documents by uploading them from your local machine or from Object Storage Service (OSS). If you upload from a local machine, you must manually update the knowledge base later. If you import from OSS, you can integrate OSS, Function Compute (FC), and the Alibaba Cloud Model Studio knowledge base API to enable automatic data updates. You can upload local files using the console or an API. You can only import from OSS using the console.For more information, see Knowledge base: Step 1. Import data.
FAQ	Create a structured knowledge base to manage FAQ-style knowledge. You can create a structured knowledge base by uploading local documents or using ApsaraDB RDS. If you upload local documents, you must manually update the knowledge base later. If you use ApsaraDB RDS, data updates in the RDS table are automatically synchronized to the knowledge base. Currently, both methods can only be performed in the console.For more information, see Knowledge base: Step 1. Import data.

Enterprise knowledge type

Migration plan

Documents

Create an unstructured knowledge base to manage document-based knowledge. Import documents by uploading them from your local machine or from Object Storage Service (OSS).

If you upload from a local machine, you must manually update the knowledge base later. If you import from OSS, you can integrate OSS, Function Compute (FC), and the Alibaba Cloud Model Studio knowledge base API to enable automatic data updates.

You can upload local files using the console or an API. You can only import from OSS using the console.For more information, see Knowledge base: Step 1. Import data.

FAQ

Create a structured knowledge base to manage FAQ-style knowledge. You can create a structured knowledge base by uploading local documents or using ApsaraDB RDS.

If you upload local documents, you must manually update the knowledge base later. If you use ApsaraDB RDS, data updates in the RDS table are automatically synchronized to the knowledge base.

Currently, both methods can only be performed in the console.For more information, see Knowledge base: Step 1. Import data.

After you create a new knowledge base, you can associate it with your Agent Application or Workflow Application in My Applications. For more information, see Knowledge base: Step 4. Reference the knowledge base.

Use the SDK

Prerequisites

You have obtained an API key and configured it as an environment variable. For more information, see Obtain an API key and Configure an API key.
The DashScope SDK is installed. Python and Java versions are available. For more information, see Install the latest version of the SDK.
A RAG application is created and enterprise knowledge is uploaded. You can no longer create new legacy RAG applications. For more information, see Upload enterprise knowledge.
Obtain the APP_ID. You can view it on the application card on the My Applications page.

Call examples

from http import HTTPStatus
from dashscope import Application

def rag_call():
    response = Application.call(app_id='YOUR_APP_ID',
                                prompt='How is the TopP parameter passed in the API operation description?',
                                )

    if response.status_code != HTTPStatus.OK:
        print('request_id=%s, code=%s, message=%s\n' % (response.request_id, response.status_code, response.message))
    else:
        print('request_id=%s\n output=%s\n usage=%s\n' % (response.request_id, response.output, response.usage))

if __name__ == '__main__':
    rag_call()

Retrieve by tag

If your enterprise knowledge base contains a large volume of data, use tag-based retrieval to improve recall accuracy. First, add knowledge tags to your documents. For more information, see Upload enterprise knowledge. Then, obtain the tag IDs and pass them in the doc_tag_codes parameter. You can pass multiple tag IDs.

When you pass tags, the retrieval process is limited to the documents associated with those tags, not the entire knowledge base.

from http import HTTPStatus
from dashscope import Application

def rag_call_with_tags():
    response = Application.call(app_id='YOUR_APP_ID',
                                prompt='How is the TopP parameter passed in the API operation description?',
                                doc_tag_codes=['471d*******3427', '471d*******3428'],  # Specify a tag scope for retrieval.
                                doc_reference_type=Application.DocReferenceType.simple,  # The result does not include reference markers.
                                has_thoughts=True  # Enable the return of retrieval process information.
                                )

    if response.status_code != HTTPStatus.OK:
        print('request_id=%s, code=%s, message=%s\n' % (response.request_id, response.status_code, response.message))
    else:
        print('request_id=%s\n output=%s\n usage=%s\n' % (response.request_id, response.output, response.usage))

        thoughts = response.output.thoughts
        if thoughts is not None:
            for thought in thoughts:
                print('thought=%s' % thought)

if __name__ == '__main__':
    rag_call_with_tags()

Multi-turn conversation

Cloud-hosted multi-turn conversation

Alibaba Cloud Model Studio applications provide a cloud-hosted multi-turn conversation feature. Use the session_id to conduct multi-turn conversations. Alibaba Cloud automatically manages the conversation history, so you do not need to maintain it on the client side. In the following example, the first call returns a session_id. When you pass this session_id in the second call, the model service includes the session information from the first call.

The session ID is valid for one hour. The maximum number of conversation turns in the history is 50.If you pass both session_id and history, the history parameter takes precedence, and the cloud-hosted conversation is not used.

from http import HTTPStatus
from dashscope import Application

def call_with_session():
    response = Application.call(app_id='YOUR_APP_ID',
                                prompt='I want to go to Xinjiang',
                                )

    if response.status_code != HTTPStatus.OK:
        print('request_id=%s, code=%s, message=%s\n' % (response.request_id, response.status_code, response.message))
        return

    response = Application.call(app_id='your app id',
                                prompt='What tourist attractions or food are there?',
                                session_id=response.output.session_id
                                )
    if response.status_code != HTTPStatus.OK:
        print('request_id=%s, code=%s, message=%s\n' % (response.request_id, response.status_code, response.message))
    else:
        print('request_id=%s, output=%s, usage=%s\n' % (response.request_id, response.output, response.usage))

if __name__ == '__main__':
    call_with_session()

Client-side multi-turn conversation

If you maintain the conversation history on the client side, you can pass the conversation information in the history parameter.

from http import HTTPStatus
from dashscope import Application

def call_with_history():
    prompt = 'I want to go to Xinjiang'
    response = Application.call(app_id="YOUR_APP_ID",
                                prompt=prompt,
                                )

    if response.status_code != HTTPStatus.OK:
        print('request_id=%s, code=%s, message=%s\n' % (response.request_id, response.status_code, response.message))
        return

    history = [{'user': prompt, 'bot': response.output.text}]
    response = Application.call(app_id="YOUR_APP_ID",
                                prompt='What tourist attractions or food are there?',
                                history=history
                                )
    if response.status_code != HTTPStatus.OK:
        print('request_id=%s, code=%s, message=%s\n' % (response.request_id, response.status_code, response.message))
    else:
        print('request_id=%s, output=%s, usage=%s\n' % (response.request_id, response.output, response.usage))

if __name__ == '__main__':
    call_with_history()

Streaming output

To enable streaming output, add the corresponding parameter. In the Python SDK, add stream=True. In the Java SDK, use the streamCall interface.

from http import HTTPStatus
from dashscope import Application

def call_with_stream():
    responses = Application.call(app_id='YOUR_APP_ID',
                                 prompt='How to make scrambled eggs with tomatoes?',
                                 stream=True
                                 )

    for response in responses:
        if response.status_code != HTTPStatus.OK:
            print('request_id=%s, code=%s, message=%s\n' % (
                response.request_id, response.status_code, response.message))
        else:
            print('output=%s, usage=%s\n' % (response.output, response.usage))

if __name__ == '__main__':
    call_with_stream()

Workspaces

The preceding examples call an application in the default workspace. To call an application in a different workspace, you must pass the specific workspace ID.

Replace YOUR_WORKSPACE in the example with your actual workspace ID to ensure the code runs correctly. To obtain the workspace ID, see Obtain a workspace ID.

from http import HTTPStatus
from dashscope import Application

def call_with_workspace():
    response = Application.call(app_id='YOUR_APP_ID',
                                workspace='YOUR_WORKSPACE',
                                prompt='How to make scrambled eggs with tomatoes?',
                                )

    if response.status_code != HTTPStatus.OK:
        print('request_id=%s, code=%s, message=%s\n' % (response.request_id, response.status_code, response.message))
    else:
        print('request_id=%s, text=%s, usage=%s\n' % (response.request_id, response.output.text, response.usage))

if __name__ == '__main__':
    call_with_workspace()

Input parameters

Parameter	Type	Default	Description
app_id	string	-	The application ID.
prompt	string	-	The instruction that you want the model to execute. This guides the model in generating a response.
session_id	string	-	The unique ID of a conversation session. If you pass a session_id, the conversation history is recorded in the cloud. The LLM automatically includes the stored conversation history in subsequent calls. Make sure that the session_id is unique. You can use either session_id or history, but not both.
history (optional)	list[dict]	[]	The conversation history between the user and the model. Each element in the list is a turn in the conversation, formatted as {"user": "user input", "bot": "model output"}. The turns are arranged chronologically.
workspace (optional)	string	-	The workspace ID.
seed (optional)	int	1234	The random number seed used for generation. This controls the randomness of the model's output. The seed must be an unsigned 64-bit integer. The default value is 1234. When a seed is used, the model tries to produce the same or similar results, but identical results are not guaranteed for every generation.
top_p (optional)	float	0.2	The probability threshold for nucleus sampling. For example, a value of 0.2 means that only the smallest set of most-likely tokens with a cumulative probability of 0.2 or higher are kept as the candidate set. The value must be in the range of (0, 1.0). A higher value increases randomness, while a lower value increases determinism.
top_k (optional)	int	None	The size of the candidate set for sampling. For example, a value of 50 means that only the 50 tokens with the highest scores are included in the candidate set for random sampling. A higher value increases randomness, while a lower value increases determinism. If this parameter is not passed, is set to None, or is greater than 100, the top_k policy is not used, and only the top_p policy is in effect.
temperature (optional)	float	1.0	Controls the degree of randomness and diversity. The temperature value smooths the probability distribution of candidate words. A higher temperature value flattens the distribution, allowing less probable words to be selected and making the output more diverse. A lower temperature value sharpens the distribution, making more probable words more likely to be selected and making the output more deterministic. Valid values: [0, 2). A value of 0 is not recommended because it is meaningless. python version >=1.10.1 java version >= 2.5.1
stream (optional)	bool	False	Specifies whether to use streaming output. When streaming is enabled, the API operation returns a generator. You must iterate through the generator to get the results. By default, each output is the entire sequence generated so far, and the final output is the complete result. You can set the incremental_output parameter to False to change to non-incremental output mode.
doc_tag_codes (optional)	list[str]	-	A list of document tag codes. When you pass a list of tag codes, the retrieval process is limited to the documents associated with those tags.
doc_reference_type	str	-	The reference data type for the retrieved documents. Set the value to `simple`. If you set the value to `simple`, the response does not contain reference markers.
has_thoughts	bool	False	Specifies whether to output information about the retrieval and recall process. If enabled, the response includes details about the document retrieval, recall, and model inference processes.

Output parameters

Response parameter	Type	Description	Remarks
status_code	int	A value of 200 (HTTPStatus.OK) indicates that the request was successful. Otherwise, the request failed. You can get the error code from the `code` field and the detailed error message from the `message` field. Note This field is for Python only. In Java, a failure throws an exception that contains the code and message.
request_Id	string	The ID generated by the system for this call.
code	string	The error code if the request fails. This is ignored on success. Python only
message	string	The detailed error message if the request fails. This is ignored on success. Python only
output	dict	The call result. For Qwen models, this includes the output text.
usage	dict	The metering data for this request.
output.text	string	The response generated by the model.
output.finish_reason	string	This is null during generation. If generation stops due to a stop token, the value is `stop`.
output.session_id	string	The unique ID of the conversation session.	In a multi-turn conversation, you can use this ID for session persistence.
usage.models[].model_id	string	The model invoked by the application for this call.
usage.models[].input_tokens	int	The number of tokens in the user input text.
usage.models[].output_tokens	int	The number of tokens in the response generated by the model.
output.thoughts[].throught	string	The model's thought process result.
output.thoughts[].action_type	string	The type of execution step returned by the LLM. `api`: Execute an API plugin. `response`: Return the final result.
output.thoughts[].action_name	string	The name of the action being executed, such as document retrieval or an API plugin.
output.thoughts[].action	string	The step being executed.
output.thoughts[].action_input_stream	string	The streaming result of the input parameter.
output.thoughts[].action_input	string	The input parameters for the plugin.
output.thoughts[].response	string	The result returned by the model call.
output.thoughts[].observation	string	The result returned by the retrieval or plugin.

HTTP API

Description

The application supports API calls that use the standard HTTP and HTTP Server-Sent Events (SSE) protocols. You can choose the protocol that suits your needs.

Prerequisites

The Alibaba Cloud Model Studio service is activated. For more information, see Activate the service.
An API key is created. For more information, see Obtain an API key.
A RAG application is created and enterprise knowledge is uploaded. For more information, see Build a Q&A application with no code and Upload enterprise knowledge.

Submit an API call

POST https://dashscope.aliyuncs.com/api/v1/apps/{YOUR_APP_ID}/completion

Request parameters

Parameter-passing method	Field	Type	Required	Description	Example
Header	Content-Type	String	Yes	The request type: application/json	application/json
	Accept	String	No	/. Set to `text/event-stream` to enable SSE responses. The default is not set.	text/event-stream
	Authorization	String	Yes	The API key. Example: Bearer d1**2a	Bearer d1**2a
	X-DashScope-SSE	String	No	Use either this or `Accept: text/event-stream` to enable SSE responses.	enable
	X-DashScope-WorkSpace	String	No	The workspace ID.	ws_ik******RVYCKzt
Body	input.prompt	String	Yes	The instruction that you want the model to execute. Both Chinese and English are supported.	Which park is closer to me?
	input.history	List	No	The conversation history between the user and the model. Each element in the list is a turn in the conversation, formatted as {"user": "user input", "bot": "model output"}. The turns are arranged chronologically.	"history": [ { "user":"How is the weather today?", "bot":"The weather is nice today. Do you want to go out?" }, { "user":"Do you have any recommendations?", "bot":"I suggest going to the park. Spring is here, and the flowers are beautiful." } ]
	input.session_id	String	No	The unique ID of a conversation session. If you pass a session_id, the conversation history is recorded in the cloud. The LLM automatically includes the stored conversation history in subsequent calls. Make sure that the session_id is unique. You can use either session_id or history, but not both.
	input.doc_tag_codes	String	No	A list of document tag codes. When you pass a list of tag codes, the retrieval process is limited to the documents associated with those tags.	["471d*****3427", "881f***0c232"]
	parameters.seed	Integer	No	The random number seed used for generation. This controls the randomness of the model's output. The seed must be an unsigned 64-bit integer. The default value is 1234. When a seed is used, the model tries to produce the same or similar results, but identical results are not guaranteed for every generation.	65535
	parameters.top_p	Float	No	The probability threshold for nucleus sampling. For example, a value of 0.2 means that only the tokens with a cumulative probability of 0.2 or higher are kept as the candidate set for random sampling. The value must be in the range of (0, 1.0). A higher value increases randomness, while a lower value decreases randomness. The default value is 0.2. Do not set the value to 1 or higher.	0.2
	parameters.top_k	Integer	No	The size of the candidate set for sampling. For example, a value of 50 means that only the 50 tokens with the highest scores are included in the candidate set for random sampling. A higher value increases randomness, while a lower value increases determinism. Note: If the top_k parameter is empty or its value is greater than 100, the top_k policy is not used, and only the top_p policy is in effect. The default is empty.	50
	parameters.temperature	Float	No	Controls the degree of randomness and diversity. The temperature value smooths the probability distribution of candidate words. A higher temperature value flattens the distribution, allowing less probable words to be selected and making the output more diverse. A lower temperature value sharpens the distribution, making more probable words more likely to be selected and making the output more deterministic. Valid values: [0, 2). The system default is 1.0. A value of 0 is not recommended because it is meaningless.	1.0
	parameters.has_thoughts	Bool	No	Specifies whether to output information about the retrieval and recall process. If enabled, the response includes details about the document retrieval, recall, and model inference processes.

Response parameters

Field	Type	Description	Example
status_code	int	A value of 200 (HTTPStatus.OK) indicates that the request was successful. Otherwise, the request failed. You can get the error code from the `code` field and the detailed error message from the `message` field. Note This field is for Python only. In Java, a failure throws an exception that contains the code and message.	200
request_Id	string	The ID generated by the system for this call.	33dcf25a-******-8b711f15614e
code	string	The error code if the request fails. This is ignored on success.
message	string	The detailed error message if the request fails. This is ignored on success.
output.text	string	The response generated by the model.
output.finish_reason	string	This is null during generation. If generation stops due to a stop token, the value is `stop`.
output.session_id	string	The unique ID of the conversation session.	In a multi-turn conversation, you can use this ID for session persistence.
usage.models[].model_id	string	The model invoked by the application for this call.
usage.models[].input_tokens	int	The number of tokens in the user input text.
usage.models[].output_tokens	int	The number of tokens in the response generated by the model.
output.thoughts[].throught	string	The model's thought process result.
output.thoughts[].action_type	string	The type of execution step returned by the LLM. `api`: Execute an API plugin. `response`: Return the final result.
output.thoughts[].action_name	string	The name of the action being executed, such as document retrieval or an API plugin.
output.thoughts[].action	string	The step being executed.
output.thoughts[].action_input_stream	string	The streaming result of the input parameter.
output.thoughts[].action_input	string	The input parameters for the plugin.
output.thoughts[].response	string	The result returned by the model call.
output.thoughts[].observation	string	The result returned by the retrieval or plugin.

Request example (SSE disabled)

The following example shows a script that uses a cURL command to call a RAG application with SSE disabled.

Replace YOUR_API_KEY with your API key and YOUR_APP_ID with your app ID to ensure the code runs correctly.

curl --location 'https://dashscope.aliyuncs.com/api/v1/apps/{YOUR_APP_ID}/completion' \
--header 'Authorization: Bearer {YOUR_API_KEY}' \
--header 'Content-Type: application/json' \
--data '{
    "input": {
        "prompt": "How to cook braised pig's feet with potatoes?"
    },
    "parameters":  {},
    "debug": {}
}' --verbose

Response example (SSE disabled)

{
  "output": {
    "finish_reason": "stop",
    "session_id": "a08c1b4e******0c4513db9eea4",
    "text": "Braised pig's feet with potatoes is a delicious and nutritious home-style dish. Here is a simple recipe:\n\n**Ingredients:**\n1. 500 g pig's feet (trotters)\n2. 2-3 potatoes\n3. 3-4 slices of ginger\n4. 1 scallion\n5. 2 star anises\n6. 1 small piece of cinnamon bark\n7. 2 bay leaves\n8. Cooking wine, to taste\n9. Light soy sauce and dark soy sauce, to taste\n10. Rock sugar, to taste\n11. Water, as needed\n12. Salt, to taste\n\n**Steps:**\n1. **Prepare the pig's feet**: Chop the pig's feet into small pieces, blanch them in boiling water to remove blood and impurities, then rinse with clean water and set aside.\n   \n2. **Caramelize the sugar**: Add a small amount of oil to a pot, add the rock sugar, and cook over low heat until it melts into a red, foamy liquid. Add the blanched pig's feet and stir-fry until they are evenly coated with the caramel.\n\n3. **Braise**: Add the sliced ginger, scallion sections, star anise, cinnamon, and bay leaves. Stir-fry until fragrant. Add the cooking wine, light soy sauce, and dark soy sauce for color. Then, add enough hot water to cover the pig's feet.\n\n4. **Simmer the pig's feet**: Bring to a boil over high heat, skim off any foam, then reduce the heat to medium-low and simmer for about 40 minutes until the pig's feet are tender.\n\n5. **Add the potatoes**: When the pig's feet are about 70-80% cooked, peel and chop the potatoes, add them to the pot, and continue to simmer for about 20 more minutes until the potatoes are cooked through and can be easily pierced with a chopstick.\n\n6. **Season**: Finally, add salt to taste and simmer for a few more minutes to let the flavors meld.\n\n7. **Serve**: When the sauce has thickened and both the potatoes and pig's feet are fully cooked, garnish with chopped scallions or cilantro, then turn off the heat and serve.\n\nThis is the basic recipe for braised pig's feet with potatoes. The cooking time may vary depending on the tenderness of the pig's feet and your personal preference. Adjust as needed."
  },
  "usage": {
    "models": [
      {
        "output_tokens": 456,
        "model_id": "qwen-max",
        "input_tokens": 64
      }
    ]
  },
  "request_id": "99432adc-8b15-953f-afba-fd0895a68773"
}

Request example (SSE enabled)

The following example shows a script that uses a cURL command to call a RAG application with SSE enabled.

Replace YOUR_API_KEY with your API key and YOUR_APP_ID with your app ID to ensure the code runs correctly.

curl --location 'https://dashscope.aliyuncs.com/api/v1/apps/{YOUR_APP_ID}/completion' \
--header 'Authorization: Bearer {YOUR_API_KEY}' \
--header 'Content-Type: application/json' \
--header 'X-DashScope-SSE: enable' \
--data '{
    "input": {
        "prompt": "How to cook braised pig's feet with potatoes?"
    },
    "parameters":  {},
    "debug": {}
}' --verbose

Response example (SSE enabled)

id:1
event:result
:HTTP_STATUS/200
data:{"output":{"session_id":"2c502d4df28******f00488c10da4","finish_reason":"null","text":"potato"},"usage":{"models":[{"input_tokens":64,"output_tokens":1,"model_id":"qwen-max"}]},"request_id":"d9abe8f8-5be6-9118-9232-8c27b9ff536c"}

id:2
event:result
:HTTP_STATUS/200
data:{"output":{"session_id":"2c502d4df28******f00488c10da4","finish_reason":"null","text":"potato stew"},"usage":{"models":[{"input_tokens":64,"output_tokens":2,"model_id":"qwen-max"}]},"request_id":"d9abe8f8-5be6-9118-9232-8c27b9ff536c"}

id:3
event:result
:HTTP_STATUS/200
data:{"output":{"session_id":"2c502d4df28******f00488c10da4","finish_reason":"null","text":"potato stew pork"},"usage":{"models":[{"input_tokens":64,"output_tokens":3,"model_id":"qwen-max"}]},"request_id":"d9abe8f8-5be6-9118-9232-8c27b9ff536c"}

... ... ... ...
... ... ... ...

id:7
event:result
:HTTP_STATUS/200
data:{"output":{"session_id":"2c502d4df28******f00488c10da4","finish_reason":"null","text":"Braised pork trotters with potatoes is a delicious and nutritious home-style dish. Below is a simple recipe:\n\n**Ingredients:**\n1. Pork"},"usage":{"models":[{"input_tokens":64,"output_tokens":32,"model_id":"qwen-max"}]},"request_id":"d9abe8f8-5be6-9118-9232-8c27b9ff536c"}

Request example (workspace)

The following example shows a script that uses a cURL command to call a RAG application in a specified workspace.

Replace YOUR_API_KEY with your API key, YOUR_APP_ID with your app ID, and YOUR_WORKSPACE with your workspace ID to ensure the code runs correctly.

curl --location 'https://dashscope.aliyuncs.com/api/v1/apps/{YOUR_APP_ID}/completion' \
--header 'Authorization: Bearer {YOUR_API_KEY}' \
--header 'Content-Type: application/json' \
--header 'X-DashScope-WorkSpace: {YOUR_WORKSPACE}' \
--data '{
    "input": {
        "prompt": "How to cook braised pig's feet with potatoes?"
    },
    "parameters":  {},
    "debug": {}
}' --verbose

Response example (workspace)

{
  "output": {
    "finish_reason": "stop",
    "session_id": "a08c1b4e******0c4513db9eea4",
    "text": "Braised pig's feet with potatoes is a delicious and nutritious home-style dish. Here is a simple recipe:\n\n**Ingredients:**\n1. 500 g pig's feet (trotters)\n2. 2-3 potatoes\n3. 3-4 slices of ginger\n4. 1 scallion\n5. 2 star anises\n6. 1 small piece of cinnamon bark\n7. 2 bay leaves\n8. Cooking wine, to taste\n9. Light soy sauce and dark soy sauce, to taste\n10. Rock sugar, to taste\n11. Water, as needed\n12. Salt, to taste\n\n**Steps:**\n1. **Prepare the pig's feet**: Chop the pig's feet into small pieces, blanch them in boiling water to remove blood and impurities, then rinse with clean water and set aside.\n   \n2. **Caramelize the sugar**: Add a small amount of oil to a pot, add the rock sugar, and cook over low heat until it melts into a red, foamy liquid. Add the blanched pig's feet and stir-fry until they are evenly coated with the caramel.\n\n3. **Braise**: Add the sliced ginger, scallion sections, star anise, cinnamon, and bay leaves. Stir-fry until fragrant. Add the cooking wine, light soy sauce, and dark soy sauce for color. Then, add enough hot water to cover the pig's feet.\n\n4. **Simmer the pig's feet**: Bring to a boil over high heat, skim off any foam, then reduce the heat to medium-low and simmer for about 40 minutes until the pig's feet are tender.\n\n5. **Add the potatoes**: When the pig's feet are about 70-80% cooked, peel and chop the potatoes, add them to the pot, and continue to simmer for about 20 more minutes until the potatoes are cooked through and can be easily pierced with a chopstick.\n\n6. **Season**: Finally, add salt to taste and simmer for a few more minutes to let the flavors meld.\n\n7. **Serve**: When the sauce has thickened and both the potatoes and pig's feet are fully cooked, garnish with chopped scallions or cilantro, then turn off the heat and serve.\n\nThis is the basic recipe for braised pig's feet with potatoes. The cooking time may vary depending on the tenderness of the pig's feet and your personal preference. Adjust as needed."
  },
  "usage": {
    "models": [
      {
        "output_tokens": 456,
        "model_id": "qwen-max",
        "input_tokens": 64
      }
    ]
  },
  "request_id": "99432adc-8b15-953f-afba-fd0895a68773"
}

Error response example

If an error occurs during the request, the output includes a code and a message that indicate the cause of the error.

{"code":"InvalidApiKey","message":"Invalid API-key provided.","request_id":"2637fcf9-32b1-9f4e-b0e9-1724d4aea00e"}

Status codes

For more information about the status codes returned by the service invocation, see Error messages.

​Migration plan

​Use the SDK

​Prerequisites

​Call examples

​Retrieve by tag

​Multi-turn conversation

​Cloud-hosted multi-turn conversation

​Client-side multi-turn conversation

​Streaming output

​Workspaces

​Input parameters

​Output parameters

​HTTP API

​Description

​Prerequisites

​Submit an API call

​Request parameters

​Response parameters

​Request example (SSE disabled)

​Response example (SSE disabled)

​Request example (SSE enabled)

​Response example (SSE enabled)

​Request example (workspace)

​Response example (workspace)

​Error response example

​Status codes

Migration plan

Use the SDK

Prerequisites

Call examples

Retrieve by tag

Multi-turn conversation

Cloud-hosted multi-turn conversation

Client-side multi-turn conversation

Streaming output

Workspaces

Input parameters

Output parameters

HTTP API

Description

Prerequisites

Submit an API call

Request parameters

Response parameters

Request example (SSE disabled)

Response example (SSE disabled)

Request example (SSE enabled)

Response example (SSE enabled)

Request example (workspace)

Response example (workspace)

Error response example

Status codes