tencent cloud

TencentDB for PostgreSQL

DokumentasiTencentDB for PostgreSQLAI CapabilitiesAI Large Model InvocationIntroduction to the tencentdb_ai Plugin Features

Introduction to the tencentdb_ai Plugin Features

Unduh
Mode fokus
Ukuran font
Terakhir diperbarui: 2026-06-08 10:32:12
Currently, both foundational large models and AI application development are advancing rapidly. TencentDB for PostgreSQL provides the tencentdb_ai plugin, which enables you to easily call accessible large model APIs within TencentDB for PostgreSQL instances to complete application development for various scenarios. This plugin supports instances with the major version PostgreSQL 17.
Note:
The prerequisite for using tencentdb_ai is that you have already created an instance with the major version PostgreSQL 17.
Creating the tencentdb_ai plugin will also create the pgcrypto plugin. Please be aware of this.

tencentdb_ai Plugin Function Description

Add a model

The add_model function is defined as follows:
tencentdb_ai.add_model(
    model_name NAME,
    version TEXT,
    region TEXT,
    json_path JSONPATH
);
The parameters are explained as follows:
model_name: the model name, such as hunyuan-lite, deepseek-r1, and so on
version: the public parameter version for this model. If the model does not require the version parameter, you can leave it blank.
region: the public parameter region for this model. If the model does not require the region parameter, you can leave it blank.
json_path: specifies how to extract the model's JSON response when the model is called. This is equivalent to applying the jsonb_path_query_first function to the model output. If this parameter is not specified, the model's raw response is output as-is by default.

Viewing Registered Models

The list_models function is defined as follows:
tencentdb_ai.list_models(void);

Updating a Model

The update_model_attr function is defined as follows:
tencentdb_ai.update_model_attr(
    model_name NAME,
    attr_name text,
    attr_value text
);
The parameters are explained as follows:
model_name: the model name.
attr_name: the model attribute to be updated. Updatable model attributes include version, region, json_path, SecretId, and SecretKey.
attr_value: the value of the attribute to be updated.
For specific SecretId and SecretKey, refer to CAM.

Delete a model

The delete_model function is defined as follows:
tencentdb_ai.delete_model(
    model_name NAME
);
The parameters are explained as follows:
model_name: the model name.

Model Invocation

The call_model function is defined as follows:
tencentdb_ai.call_model(
    model_name NAME,
    common_params TEXT[],
    api_params TEXT[]
);
The parameters are explained as follows:
model_name: the model name.
common_params: the public parameters required to invoke the model, for example, ARRAY['Action: ChatCompletions', 'Version: 2023-09-01']
api_params: the proprietary parameters required to invoke the model, for example, ARRAY['"Stream": false', '"Model": "hunyuan-lite"', '"Messages": [{"Role": "user", "Content": "Hello"}]']
In addition to using the call_model function to invoke models, you can also use the following APIs for model invocation in specific scenarios.

Fixed-Scene APIs

Dialogue

The chat_completions function is defined as follows:
tencentdb_ai.chat_completions(
    model_name  NAME,
    content TEXT,
    args TEXT[] default NULL
);
The parameters are explained as follows:
model_name: the model name.
content: the conversation content.
args: other parameters for model invocation, with a default value of NULL.
Invocation example:
SELECT tencentdb_ai.chat_completions('hunyuan-lite', 'Hello');
                                                                                                                                                                                          
         chat_completions                                                                                                                                                                 
                                  
------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
----------------------------------
 {"Response":{"RequestId":"*******-****-****-****-***********","Note":"The above content is AI-generated and does not represent the developer's position. Please do not delete or modify this marker.","Choices":[{"Index":0,"Message":{"Role":"assistant","Co
ntent":"Hello there! I'm very happy to chat with you. Is there anything you'd like to share with me today?"},"FinishReason":"stop"}],"Created":1739786393,"Id":"*******-****-****-****-*********","Usage":{"PromptTokens":3,"Completio
nTokens":17,"TotalTokens":20}}}
(1 row)

Text to Vector

The get_embedding function is defined as follows:
tencentdb_ai.get_embedding(
    model_name NAME, 
    content TEXT[]
);
The parameters are explained as follows:
model_name: the model name.
content: the content to be transformed.
Invocation example:
damoxing1=> SELECT tencentdb_ai.get_embedding('hunyuan-embedding', ARRAY['Hello', 'PostgreSQL']);

Text Ranking

The run_rerank function is defined as follows:
tencentdb_ai.run_rerank(
    model_name NAME, 
    query TEXT, 
    documents TEXT[], 
    args TEXT[] DEFAULT NULL
);
The parameters are explained as follows:
model_name: the model name.
query: the text sorting problem.
documents: the content to be sorted by text.
args: the specified parameters for a specific model.
damoxing1=> SELECT COUNT(*) FROM tencentdb_ai.run_rerank('lke-reranker-base', 'Knowledge Engine Large Model', ARRAY['Hunyuan Large Model', 'Tencent Knowledge Engine']);
 count 
-------
     2
(1 row)

Image to Text

The image_to_text function is defined as follows:
tencentdb_ai.image_to_text
(
    model_name NAME, 
    query TEXT, 
    image TEXT, 
    args TEXT[] DEFAULT NULL
);
The parameters are explained as follows:
model_name: the model name.
query: the image-to-text problem.
image: the image URL.
args: the specified parameters for a specific model.

Plugin Usage Examples

The process of using the extension is described in detail below:
1. Creating Extension
damoxing=> CREATE EXTENSION tencentdb_ai CASCADE;
NOTICE:  installing required extension "pgcrypto"
CREATE EXTENSION
Check that the extension has been created.
damoxing=> SELECT * FROM pg_extension;
  oid  |   extname    | extowner | extnamespace | extrelocatable | extversion | extconfig | extcondition 
-------+--------------+----------+--------------+----------------+------------+-----------+--------------
 14275 | plpgsql      |       10 |           11 | f              | 1.0        |           | 
 16440 | pgcrypto     |    16437 |         2200 | t              | 1.3        |           | 
 16477 | tencentdb_ai |    16437 |         2200 | t              | 1.0        | {16479}   | {""}
(3 rows)
2. Add a large model.
damoxing=> SELECT tencentdb_ai.add_model('hunyuan-lite','2023-09-01', NULL, NULL);
 add_model 
-----------
 
(1 row)
3. Add the SecretId and SecretKey for large model API calls.
For specific SecretId and SecretKey, refer to CAM.
damoxing=> SELECT tencentdb_ai.update_model_attr('hunyuan-lite', 'SecretId', 'AKID***************');
 update_model_attr 
-------------------
 
(1 row)
damoxing=> SELECT tencentdb_ai.update_model_attr('hunyuan-lite', 'SecretKey', '********************');
 update_model_attr 
-------------------
 
(1 row)
4. View the list of currently added models.
damoxing=> SELECT * FROM tencentdb_ai.list_models();
-[ RECORD 1 ]--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
model_name | hunyuan-lite
json_path  | 
secretid   | ***************************
secretkey  | ***************************
version    | 
region     | 
id_random  | 516234453022
key_random | 134577899689
5. Call the large model.
For chat scenarios, you can use the call_model API.
damoxing=> SELECT tencentdb_ai.call_model('hunyuan-lite',
                            ARRAY['Action: ChatCompletions', 'Version: 2023-09-01'],
                            ARRAY['"Stream": false', '"Model": "hunyuan-lite"', '"Messages": [{"Role": "user", "Content": "Hello"}]']);damoxing(> damoxing(> 
                                                                                                                                      
                                                                                                  call_model                          
                                                                                                                                      
                                                                        
--------------------------------------------------------------------------------------------------------------------------------------
--------------------------------------------------------------------------------------------------------------------------------------
--------------------------------------------------------------------------------------------------------------------------------------
------------------------------------------------------------------------
 {"Response":{"RequestId":"*******-****-****-****-***********","Note":"The above content is AI-generated and does not represent the developer's position. Please do not delete or modify this marker.","Ch
oices":[{"Index":0,"Message":{"Role":"assistant","Content":"Hello! I'm very happy to communicate with you. Is there anything I can help you with? Whether it's about life, work, or study,
or any other questions, I will do my best to help you."},"FinishReason":"stop"}],"Created":1739793838,"Id":"*******-****-****-****-***********
","Usage":{"PromptTokens":3,"CompletionTokens":33,"TotalTokens":36}}}
(1 row)

damoxing=>
Topik sebelumnya: Overview of AI Capabilities for TencentDB for PostgreSQLTopik selanjutnya: Building Applications by Calling Large Models Using tencentdb_ai

Bantuan dan Dukungan

Apakah halaman ini membantu?

masukan