It supports major features of Vectara's RAG platform for you to build your own search engine.
Additional features:
- It expands the upload function to allow you to upload a list of files or all files under a directory in one function call.
- Renders query results in beautiful Markdown printout (see
demo.ipynb).
# Stable
pip install vectara
# Nightly
pip install "git https://github.com/forrestbao/vectara-python-cli.git"You can obtain the Vectara credentials following this guide. Then set up your Vectara credentials as environment variables.
export VECTARA_CUSTOMER_ID=123
export VECTARA_CLIENT_ID=abc
export VECTARA_CLIENT_SECRET=xyzThis unofficial SDK and CLI will read your credentials from the environment variables above. The ability to pass in credentials as arguments is also supported.
Try the Jupyter notebook demo.ipynb or read the docstring.
from vectara import vectara
client = vectara() # get default credentials from environment variables
# You can also manually pass in your credentials as arguments
corpus_id = client.create_corpus('my knowledge base')
client.upload(corpus_id, 'one_file.pdf', description='My precious doc') # add one file to the corpus
client.upload(corpus_id, 'a_folder_of_documents') # add all files under a folder to the corpus
client.upload(corpus_id, ['user_manual.md', 'notes.txt'], description=['user manual', 'my memory']) # add a list of files to the corpus
client.query(corpus_id, 'Vectara allows me to search for anything, right?', top_k=5) # query the corpus for top 5 answers
client.reset_corpus(corpus_id) # delete all documents in the corpusTo learn the command line usage, run vectara --help.
You must set up your Vectara credentials as environment variables before using the command line interface.
# create a corpus
vectara create_corpus 'my knowledge base'
# output: corpus_id = 12
# upload a file to the corpus
vectara upload 12 one_file.pdf # corpurs_id = 12
# upload a folder to the corpus
vectara upload 12 ./a_folder_of_documents # corpurs_id = 12
# query the corpus
vectara query 12 'Vectara allows me to search for anything, right?' --top_k=5 # corpurs_id = 12
# reset the corpus
vectara reset_corpus 12 # corpurs_id = 12The query results are typeset Markdown ready to be rendered. A query result includes the following info:
- A summary with citations and matching scores with respect to the query
- References cited by the summary
### Here is the answer
To rearrange objects, you can utilize the "direction" attribute in a Funix decorator [1]. Manually resizing and positioning objects can be a tedious and inefficient process [2]. Another approach is to use a collision-free algorithm for auto-layout, where scopes will be resized to fit the objects inside [4]. An example of arranging objects in a column-reverse direction can be seen in the ChatGPT multiturn app [3]. Additionally, organizing your canvas with scopes can help in rearranging objects effectively [5]. Remember to experiment with these methods to find the best arrangement for your specific needs.
### References:
1. From document **funix.md** (matchness=0.673933):
_...You can change their order and orientation using the "direction" attribute in a Funix decorator...._
2. From document **codepod.md** (matchness=0.65305215):
_...It is painful and inefficient to resize and position the pods and scopes manually...._
3. From document **funix.md** (matchness=0.6520513):
_...The example below shall be self-explaining:
A more advanced example is our ChatGPT multiturn app where "direction = "column-reverse"" so the message you type stays at the bottom...._
4. From document **codepod.md** (matchness=0.6495899):
_...default}alt="Example banner"width="600"/>
After auto-layout, the pods and scopes are organized by a collision-free algorithm, and the scopes will be resized to fit the pods inside...._
5. From document **codepod.md** (matchness=0.6460015):
_...Organize your Canvas with scopes..._Contact forrest at vectara dot com
This is an UNofficial SDK and CLI for Vectara's RAG platform. Use at your own risk. Vectara does NOT provide support for this SDK or CLI.