Movatterモバイル変換

[0]ホーム
URL:

Skip to content

Navigation Menu

Sign in
Appearance settings

Search code, repositories, users, issues, pull requests...

Provide feedback

We read every piece of feedback, and take your input very seriously.

Saved searches

Use saved searches to filter your results more quickly

 Sign up
Appearance settings

A Python SDK for accessing the Vectara API

NotificationsYou must be signed in to change notification settings

vectara/python-sdk

Repository files navigation

fern shieldpypi

The Vectara Python SDK provides convenient access to the Vectara API for building powerful AI applications.

Installation

Install the library via pip:

pip install vectara

Getting Started

API Generated Documentation

API reference documentation is availablehere.

Examples

Complete examples can be found in theGetting Started notebooks.

Usage

First, create an SDK client.
You can use either anapi_key or OAuth (client_id andclient_secret) forauthentication.

fromvectaraimportVectara# creating the client using API keyclient=Vectara(api_key="YOUR_API_KEY")# creating the client using oauth credentialsclient=Vectara(client_id="YOUR_CLIENT_ID",client_secret="YOUR_CLIENT_SECRET",)

If you don't already have a corpus, you can create it using the SDK:

client.corpora.create(name="my-corpus",key="my-corpus-key")

Add a document to a corpus

You can add documents to a corpus in two formats:structured orcore.
For more information, refer to theIndexing Guide.

Here is an example for adding a Structured document

fromvectaraimportStructuredDocument,StructuredDocumentSectionclient.documents.create(corpus_key="my-corpus-key",request=StructuredDocument(id="my-doc-id",type="structured",sections=[StructuredDocumentSection(id="id_1",title="A nice title.",text="I'm a nice document section.",metadata={'section':'1.1'}          ),StructuredDocumentSection(id="id_2",title="Another nice title.",text="I'm another document section on something else.",metadata={'section':'1.2'}          ),        ],metadata={'url':'https://example.com'}    ),)

And here is one with Core document:

fromvectaraimportCoreDocument,CoreDocumentPartclient.documents.create(corpus_key="my-corpus-key",request=CoreDocument(id="my-doc-id",type="core",document_parts=[CoreDocumentPart(text="I'm a first document part.",metadata={'author':'Ofer'}            )CoreDocumentPart(text="I'm a second document part.",metadata={'author':'Adeel'}            )        ],metadata={'url':'https://example.com'}    ),)

Upload a file to the corpus

In addition to creating a document as shown above (using StructuredDocument or CoreDocument), you can also upload files (such as PDFs or Word Documents) directly to Vectara.In this case Vectara will parse the files automatically, extract text and metadata, chunk them and add them to the corpus.

Using the SDK you need to provide both the file name, the binary content of the file, and the content_type, as follows:

filename="examples.pdf"withopen(filename,"rb")asf:content=f.read()client.upload.file('my-corpus-key',file=content,filename=filename,metadata={"author":"Adeel"})

Querying the corpora

With the SDK it's super easy to run a query from one or more corpora. For more detailed information, see thisQuery API guide

A query uses two important objects:

  • TheSearchCorporaParameters object defines parameters for search such as hybrid search, metadata filtering or reranking
  • TheGenerationParameters object defines parameters for the generative step.

Here is an example query for our corpus above:

search=SearchCorporaParameters(corpora=[KeyedSearchCorpus(corpus_key="my-corpus-key",metadata_filter="",lexical_interpolation=0.005,            )        ],context_configuration=ContextConfiguration(sentences_before=2,sentences_after=2,        ),reranker=CustomerSpecificReranker(reranker_id="rnk_272725719"        ),    )generation=GenerationParameters(response_language="eng",enable_factual_consistency_score=True,    )client.query(query="Am I allowed to bring pets to work?",search=search,generation=generation    )

Using Chat

Vectarachat provides a way to automatically store chat history to support multi-turn conversations.

Here is an example of how to start a chat with the SDK:

fromvectaraimportSearchCorporaParameterssearch=SearchCorporaParameters(corpora=[KeyedSearchCorpus(corpus_key="test-corpus",metadata_filter="",lexical_interpolation=0.005,            )        ],context_configuration=ContextConfiguration(sentences_before=2,sentences_after=2,        ),reranker=CustomerSpecificReranker(reranker_id="rnk_272725719"        ),    )generation=GenerationParameters(response_language="eng",citations=CitationParameters(style="none",        ),enable_factual_consistency_score=True,    )chat=ChatParameters(store=True)session=client.create_chat_session(search=search,generation=generation,chat_config=chat,)response_1=session.chat(query="Tell me about machine learning.")print(response_1.answer)response_2=session.chat(query="what is generative AI?")print(response_2.answer)

Note that we used thecreate_chat_session withchat_config set for storing chat history. The resulting session can then be used for turn-by-turn chat, simply by using thechat() method of the session object.

Streaming

The SDK supports streaming responses for both query and chat. When using streaming, the response will be a generator that you can iterate.

Here's an example of callingquery_stream:

Streaming the query response

fromvectaraimportSearchCorporaParameterssearch=SearchCorporaParameters(corpora=[...],    ...)generation=GenerationParameters(...)response=client.query_stream(query="Am I allowed to bring pets to work?",search=search,generation=generation    )forchunkinresponse:ifchunk.type=='generation_chunk':print(chunk.generation_chunk)ifchunk.type=="search_results":print(chunk.search_results)

And streaming the chat response:

fromvectaraimportSearchCorporaParameterssearch=SearchCorporaParameters(corpora=[...],    ...)generation=GenerationParameters(...)chat_params=ChatParameters(store=True)session=client.create_chat_session(search=search_params,generation=generation_params,chat_config=chat_params,)response=session.chat_stream(query="Tell me about machine learning.")forchunkinresponse:ifchunk.type=='generation_chunk':print(chunk.generation_chunk)ifchunk.type=="search_results":print(chunk.search_results)ifchunk.type=="chat_info":print(chunk.chat_id)print(chunk.turn_id)

Additional Functionality

There is a lot more functionality packed into the SDK, matchingall API endpoints that are available in Vectara including for things like managing documents, corpora, api keys, users, and even for query history retrieval.

Exception Handling

When the API returns a non-success status code (4xx or 5xx response), a subclass of the following errorwill be thrown.

fromvectara.core.api_errorimportApiErrortry:client.query(...)exceptApiErrorase:print(e.status_code)print(e.body)

Pagination

Paginated requests will return aSyncPager orAsyncPager, which can be used as generators for the underlying object.

response=client.corpora.list(limit=1,)foriteminresponse:yielditem# alternatively, you can paginate page-by-pageforpageinresponse.iter_pages():yieldpage

Advance Usage

For more information related to customization, Timeouts and Retries in the SDK, refer to theAdvanced Usage Guide

Using the SDK in Different Contexts

The Python library can be used in a number of environments with different requirements:

  1. Notebooks - using implicit configuration from a users home directory
  2. Docker Environments - using ENV variables for configuration
  3. Complex Applications - allowing explicit configuration from mutable stores (e.g. RDBMS / NoSQL)

For more details, refer to theConfiguration Guide

Author

👤Vectara

🤝 Contributing

Contributions, issues and feature requests are welcome!
Feel free to checkissues page. You can also take a look at thecontributing guide.

Show your support

Give a ⭐️ if this project helped you!

About

A Python SDK for accessing the Vectara API

Topics

Resources

Contributing

Stars

Watchers

Forks

Contributors9

Languages

[8]ページ先頭
©2009-2025 Movatter.jp