This API is used to create a knowledge base.

X-Auth-Token

Yes

String

Parameter description:

Token used for API authentication. For how to obtain the token, see section 3.2 "Authentication."

Constraints:

N/A.

name

Yes

String

Parameter description:

Knowledge base name

Constraints:

The parameter value consists of 1 to 64 characters, including letters, digits, underscores (_), and hyphens (-).

Default value:

N/A.

detail

Yes

String

Parameter description:

Description of the knowledge base.

Constraints:

The value contains 1 to 100 characters.

Default value:

N/A.

embedding_model

Yes

String

Parameter description:

Name of the embedding model used by the current knowledge base.

Embedding model: a model used to vectorize text content for vector retrieval.

Constraints:

• The parameter value consists of 1 to 32 characters, including letters, digits, underscores (_), and hyphens (-).

• The value must be an embedding model configured on the platform. You can check the models configured on the platform using the ListModels API.

Default value:

N/A.

rerank_model

Yes

String

Parameter description:

Name of the reranking model used by the current knowledge base.

Reranking model: Reranks the top_k results returned by the embedding model through vector search. The purpose is to provide results that are most relevant to the query.

Constraints:

• The parameter value consists of 1 to 32 characters, including letters, digits, underscores (_), and hyphens (-).

• The reranking model must be configured on the platform. You can check the models configured on the platform using the ListModels API.

Default value:

N/A.

pangu_nlp_model

Yes

String

Parameter description:

Name of the NLP foundation model used by the current knowledge base.

NLP model: a generative model used by the chat API to generate answers based on the input text.

Constraints:

• The parameter value consists of 1 to 32 characters, including letters, digits, underscores (_), and hyphens (-).

• The value must be an NLP model configured on the platform. You can check the models configured on the platform using the ListModels API.

Default value:

N/A.

file_extract

Yes

FileExtractConf object

Overall configuration of document parsing

search_plan_category_ids

No

Array of strings

Parameter description:

Search category

If there is no hit on category, the knowledge base is searched and then an LLM summarizes the results.

If a category is matched, the LLM answers the question directly.

Value range:

["People", "Weather", "Industry knowledge", "Industry knowledge-manufacturing", "Industry knowledge-medical", "Industry knowledge-government", "Industry knowledge-finance", "Language task", "General knowledge", "Chitchat"]

Constraints:

N/A

language_id

No

String

Parameter description:

Language ID of the current knowledge base.

Options:

zh==CHINESE

en==ENGLISH

th==THAI

ar==ARABIC

Constraints:

N/A.

cache_enabled

No

Boolean

Parameter description:

Whether to enable the cache function in the current knowledge base.

false: The result is not cached.

true: The query/reference/answer result is cached. When a similar query is encountered during the next retrieval, the cached result is directly returned to improve the latency.

Constraints:

N/A.

Default value:

false

answer_reference_enabled

No

Boolean

Parameter description:

Whether to enable reference source tracing in the current knowledge base to locate the reference source of the answer.

Constraints:

N/A.

Default value:

false

answer_image_reference_enabled

No

Boolean

Parameter description:

Whether to include both text and images.

true: If a document contains images related to a question, the answer must include both text and images.

false: No image is recalled.

Constraints:

Answer reference (answer_reference_enabled) must be enabled.

Default value:

false

session_config

No

SessionConfig object

Cache policy

extend_config

No

KnowledgeRepoExtendConfig object

Extended configuration of the knowledge base.

tags

No

Array of TagInfo objects

Label list

refs

No

String

Parameter description:

List of knowledge base IDs, which are separated by commas (,). It is used to support joint retrieval of Q&As in multiple knowledge bases.

Constraints:

Use commas (,) to separate different knowledge base IDs.

search_plan_model

No

String

Parameter description:

The name of the search planning model.

search_plan model: search planning model, which replans the query entered by the user to a query that is more suitable for retrieval, including query rewriting and query completion.

Constraints:

• The value can contain 1 to 32 characters, including only Chinese characters, letters, digits, underscores (_), and hyphens (-).

• The search_plan model must be configured on the platform. You can check the models configured on the platform using the ListModels API.

prompt_info

No

KnowledgeRepoPromptInfo object

Prompt information.

parse_conf

No

ParseConf object

Parameter description:

Document parsing configuration, including whether to use OCR enhancement, whether to parse images, whether to extract text during image parsing, whether to parse the header and footer, and whether to parse the contents page.

Constraints:

N/A.

split_conf

No

SplitConf object

Parameter limitations

Split configuration, including the segmentation mode, level parsing mode, title level depth, title saving mode, segment length, and title matching pattern.

Constraints:

N/A.

id

No

String

Parameter description:

Document parsing ID.

Constraints:

N/A.

ocr_enabled

No

Boolean

Parameter description:

Whether the current knowledge base uses OCR enhancement.

• Pure Word documents do not need to be parsed using OCR.

• PDF and PPTX files require OCR for intelligent document recognition, such as table parsing and text extraction.

  Constraints:

  N/A.

Default value:

false

image_enabled

No

Boolean

Parameter description:

Whether the current knowledge base needs to parse images.

true: Skip images in the document by default.

false: Parse images. The parsing mode is configured in image_conf.

Constraints:

N/A.

Default value:

false

header_footer_enabled

No

Boolean

Parameter description:

Whether to parse the header and footer of the file in the current knowledge base.

true: The parsing result contains the header and footer.

false: The parsing result does not contain the header and footer. (If the header and footer do not contain key text information, you are advised to set this parameter to false to avoid interference.)

Constraints:

N/A

Default value:

false

catalog_enabled

No

Boolean

Parameter description:

Indicates whether to parse the directory page of the file in the current knowledge base.

false: The parsing result does not contain the directory page. (If there is no information that needs to be reserved on the content page, it is recommended that the default value be false.) Generally, a directory page contains a large number of keywords, which may affect the search result.)

true: The parsing result contains the directory page.

Constraints:

N/A.

Default value:

false

image_conf

No

String

Parameter description:

Image parsing mode when image parsing is enabled (image_enable is set to True).

• TEXT: Extracts text from an image and does not retain the image.

• IMAGE: The original image is retained.

  Constraints:

  If you want to return an answer with text and images, you must use the IMAGE mode and retain the original image.

Default value:

TEXT

split_mode

No

String

Parameter description:

Mode for splitting a document.

Options:

Four modes are available:

• AUTO: The system automatically identifies the document format and matches the appropriate splitting and parsing mode.

• LENGTH: Split by length. For example, each 500 characters are split into a paragraph.

• CATALOG: Automatic parsing in hierarchical segmentation. The system automatically identifies the hierarchical structure of an article and segments the article based on the hierarchical structure. For example, section 1.1.2 is a segment, and section 1.1.3 is a segment.

• RULE: Rule-based parsing in hierarchical segmentation. You can customize the matching rules of hierarchical titles and match and split chapters based on the customized rules.

Constraints:

N/A

Default value:

AUTO

separator_ids

No

Array of strings

Parameter description:

ID list of segment identifiers in automatic segmentation and length segmentation modes.

Segment identifier: determines the end character when a slice is segmented.

Options:

The specific value mapping is as follows:

period_zh: Chinese period.

period_en: English period.

exclamation_mark_zh: Chinese exclamation mark (!)

exclamation_mark_en: English exclamation mark (!)

question_mark_zh: Chinese question mark (?)

question_mark_en: English question mark (?)

comma_zh: Chinese comma (,)

comma_en: English comma (,)

space_en: space

Constraints:

N/A.

Default value:

["period_zh", "period_en", "exclamation_mark_zh", "exclamation_mark_en", "question_mark_zh", "question_mark_en"]

rule_regex_id

No

String

Parameter description:

ID of the selected user-defined parsing rule.

Constraints:

N/A.

chunk_size

No

Integer

Parameter description:

Maximum length of a document segment. A document is segmented based on the maximum length.

Constraints:

N/A.

Default value:

500

title_level

No

Integer

Parameter description:

Depth of the title level reserved for a segment. For example:

If the depth is 3, the current paragraph is 1.1.3, and the parent titles 1.1 and 1 are retained.

If the depth is 2, the current paragraph is 1.1.3, the parent title 1.1 is retained, and the parent title 1 is discarded.

Constraints:

N/A.

Default value:

3

combine_title

No

Boolean

Parameter description:

Whether to retain the hierarchical title combination.

The options are as follows:

false: Only the last-level title is retained.

true: Save the combination of multiple levels of titles, from the first level to the last level. For example, 1.1 indicates the usage description, and 1.1.1 indicates how to open the refrigerator.

Constraints:

N/A.

Default value:

false

merge_titles

No

Boolean

Parameter description:

Whether to merge titles.

The options are as follows:

true: If the text in a single paragraph of different titles is small, the paragraphs are automatically merged into the specified segment length to generate more comprehensive results. For example, if the two adjacent sub-paragraphs are less than 200 characters and the expected segment length is 500, the two paragraphs are combined into one paragraph.

false: Paragraphs with different titles are not merged.

Constraints:

N/A.

Default value:

true

similarity_threshold

Yes

Float

Parameter description:

Threshold of the query2query similarity for matching cached questions.

Options:

0.1 to 1.0. A higher threshold indicates a higher similarity between the query and the cached question.

Constraints:

N/A.

answer_select_policy

Yes

String

Parameter description:

Cache hit selection policy.

Options:

FIRST: Select the result with the highest score as the answer.

RANDOM: Randomly select a result as the answer.

Constraints:

N/A.

eviction

Yes

Eviction object

Cache expiration policy.

model_name

Yes

String

Parameter description:

Name of the query2query model used when the cache is hit. This parameter is used to calculate the similarity between the new query and the cached query.

Constraints:

N/A.

policy

Yes

String

Parameter description:

Declares which expiration policy is used by the cache.

Options:

LRU: (Least Recently Used) now - accessTime > ttl, clear.

FIFO: (First In First Out) now - createTime > ttl, clear.

LFU: (Least Frequency Used) hit_count < threshold, clear.

Constraints:

N/A.

ttl

No

Long

Parameter description:

Cache expiration time. When the cache exceeds the specified time, the cache is cleared. The unit is millisecond.

Constraints:

N/A.

hit_count_threshold

No

Long

Parameter description:

Cache hit threshold. When the number of cache hits reaches the threshold, the cache result is not used.

Constraints:

N/A.

extend_context

No

Boolean

Parameter description:

Specifies whether to extend the long context of the reference shard. Provides a wider context to provide the model with complete answers.

Constraints:

N/A.

effective_input_length

No

Integer

Parameter description:

Specifies the length of the selected context when the extended context is enabled.

This parameter is related to the model and ensures the valid length of the input token to ensure the optimal output.

Constraints:

For multi-round dialogs, it is recommended that the value be 60/ %(rounded up) of the model context length.

Options:

2 to 128, in KB

top_p

No

Float

Parameter description:

An alternative to temperature sampling, called nucleus sampling, which controls the diversity of generated text by limiting the range of vocabulary choices. A higher top_p value means a wider choice of tokens and hence a higher text diversity.

Constraints:

You are advised to change the value of top_p or temperature to adjust the generating text tendency. Do not change both two parameters.

Options:

0.1 ~ 1

Default value:

0.1

max_tokens

No

Integer

Parameter description:

Specifies the maximum number of new words generated by the model.

Constraints:

The value of max_tokens is related to the maximum context length supported by the model. max_tokens must be less than the maximum context length supported by the model minus the length of the tokens input to the model.

Options:

1 ~ 131072

Default value:

131072

chat_temperature

No

Float

Parameter description:

Diversity and creativity of text generated by the model in non-search enhancement scenarios.

A value close to 0 indicates the lowest randomness, while 1 indicates the highest randomness. Generally, a lower temperature is suitable for deterministic tasks, higher values favor creative tasks.

Constraints:

N/A.

Options:

0 ~ 1.

search_temperature

No

Float

Parameter description:

Diversity and creativity of text generated by the model in search enhancement scenarios.

A value close to 0 indicates the lowest randomness, while 1 indicates the highest randomness. Generally, a lower temperature is suitable for deterministic tasks, higher values favor creative tasks.

Constraints:

N/A.

Options:

0 to 1. Generally, the value is set to 0.2 or 0.3 for the Pangu NLP model.

Default value:

0.3

presence_penalty

No

Float

Parameter description:

Degree of duplication in the generated text. The purpose of presence_penalty is to reduce the repeated use of the same or similar content when the model generates text, so as to improve the diversity of the generated text. If a token has appeared in the previous text, the model will be penalized when generating this token.

A smaller presence_penalty indicates that the model considers fewer previously generated tokens, which may result in repeated content in the text.

A larger value of presence_penalty indicates that the model tends to generate new tokens that have not appeared before, and the generated text is more diversified.

Options:

The value ranges from –2 to 2. The actual value needs to be determined depending on the situation. Generally, the value 1.1 is used for the Pangu NLP model.

Default value:

0

use_system_prompt

No

Boolean

Parameter description:

Whether to use the system prompt. The prompt standard combination scheme in the RAG scenario of the Pangu NLP model is used.

Constraints:

When the Pangu NLP model is used, the system prompt can be used in common scenarios.

Default value:

false

system_prompt

No

String

Parameter description:

System prompt.

Constraints:

1. This parameter is mandatory when use_system_prompt is set to true.

2. You do not need to combine the query.

embedding_search_enable

No

Boolean

Parameter description:

Specifies whether to enable vector retrieval when related documents are retrieved.

Constraints:

N/A.

Default value:

true

keyword_search_enable

No

Boolean

Parameter description:

Specifies whether to enable keyword retrieval when related documents are retrieved.

Constraints:

N/A.

Default value:

false

keyword_top_k

No

Integer

Parameter description:

Specifies the number of top results returned when keyword retrieval is used.

Constraints:

N/A.

Options:

0 ~ 100

Default value:

10

refuse_enable

No

Boolean

Parameter description:

If no related reference document content is found, determines whether to disable model invoking and directly reject the answer on the platform.

Constraints:

N/A.

Default value:

false

refuse_answer

No

String

Parameter description:

When refuse_enabled is set to true, if no related reference document content is found, the platform rejects the configured wording.

Constraints:

N/A.

image_match_type

No

String

Parameter description:

Specifies the image recall mode in the image and text recall scenario.

Options:

Currently, only context_match and reference_match are supported.

context_match: Only semantically related images are recalled. If the context of the image in the reference paragraph is semantically similar to the generated paragraph, the image is recalled. Otherwise, the image is not recalled.

reference_match: All images in the reference paragraph are recalled.

Constraints:

N/A.

Default value:

context_match

custom_types

No

Map<String,String>

Parameter description:

Mapping type dictionary of a custom field, which applies to structured data scenarios and specifies queryable fields.

Example: {"companyName": "keyword"}

companyName: field to be retrieved; keyword: retrieval mode

Constraints:

The value in the mapping dictionary must be of a type supported by Elasticsearch queries, for example, keyword, integer, or text.

tag_key

Yes

String

Parameter description:

Tag key.

Options:

The value can contain 1 to 36 characters. Only digits, letters, hyphens (-), and underscores (_) are allowed.

Constraints:

N/A.

tag_value

Yes

String

Parameter description:

Tag value.

Options:

The value can contain 0 to 43 characters. It can only contain digits, letters, hyphens (-), and underscores (_).

Constraints:

N/A.

prompt_id

No

String

Parameter description:

ID of the NLP model prompt used by the current knowledge base.

Constraints:

The value of prompt_id must be an existing prompt_id in the prompt management.

qa_question_prompt_id

No

String

Parameter description:

ID of the QA question generation prompt used by the current knowledge base. This prompt is used to automatically generate QA pairs using documents.

Constraints:

The value of prompt_id must be an existing prompt_id in the prompt management.

qa_answer_prompt_id

No

String

Parameter description:

ID of the QA answer generation prompt used by the current knowledge base. This prompt is used to automatically generate QA pairs using documents.

Constraints:

The value of prompt_id must be an existing prompt_id in the prompt management.

repo_id

String

Parameter description:

Specifies the current knowledge base ID.

Constraints:

N/A

error_code

String

Error Code

error_msg

String

Error description

error_code

String

Error Code

error_msg

String

Error description

200

Knowledge base ID.

400

Incorrect request body parameter

500

Internal error