* Move the text rendering out for reusability
* Refactor common operations in the reasoning pipeline
* Add run method
* Provide dedicated method for invoke
* Treat index id as auto-generated field
* fix Can't create index: KeyError: 'embedding' #28
* udpate docs
* Update requirement
* Use lighter default local embedding model
---------
Co-authored-by: ian <ian@cinnamon.is>
* Allow adding indices
* Allow deleting indices
* Allow updating the indices
* When there are multiple indices, group them below Indices tab
* Update elem classes
* Provide the Embedding management UI
* Update Fastembed documentation
* Add validation when adding / updating embeddings
* Stop using the old ktem embeddings manager
* Set default local embedding models
* Move the local embeddings below in flowsettings
* Update flowsettings
* Prepend all Langchain-based embeddings with LC
* Provide vanilla OpenAI embeddings
* Add test for AzureOpenAIEmbeddings and OpenAIEmbeddings
* Fix disallowed empty string
* Use OpenAIEmbeddings in flowsettings
---------
Co-authored-by: ian_Cin <ian@cinnamon.is>
* Prepend all Langchain-based embeddings with LC
* Provide vanilla OpenAI embeddings
* Add test for AzureOpenAIEmbeddings and OpenAIEmbeddings
* Incorporate fastembed
---------
Co-authored-by: ian_Cin <ian@cinnamon.is>
* Rename AzureChatOpenAI to LCAzureChatOpenAI
* Provide vanilla ChatOpenAI and AzureChatOpenAI
* Remove the highest accuracy, lowest cost criteria
These criteria are unnecessary. The users, not pipeline creators, should choose
which LLM to use. Furthermore, it's cumbersome to input this information,
really degrades user experience.
* Remove the LLM selection in simple reasoning pipeline
* Provide a dedicated stream method to generate the output
* Return placeholder message to chat if the text is empty
* Auto create conversation when the user starts
* Add conversation rename rule check
* Fix empty name during save
* Confirm deleting conversation
* Show warning if users don't select file when upload files in the File Index
* Feedback when user uploads duplicated file
* Limit the file types
* Fix valid username
* Allow login when username with leading and trailing whitespaces
* Improve the user
* Disable admin panel for non-admnin user
* Refresh user lists after creating/deleting users
* Auto logging in
* Clear admin information upon signing out
* Fix unable to receive uploaded filename that include special characters, like !@#$%^&*().pdf
* Set upload validation for FileIndex
* Improve user management UI/UIX
* Show extraction error when indexing file
* Return selected user -1 when signing out
* Fix default supported file types in file index
* Validate changing password
* Allow the selector to contain mulitple gradio components
* A more tolerable placeholder screen
* Allow chat suggestion box
* Increase concurrency limit
* Make adobe loader optional
* Use BaseReasoning
---------
Co-authored-by: trducng <trungduc1992@gmail.com>
* Add regen button and repharasing question on regen
* Stop appending regen messages to history, allow only one
* Add dynamic conversation state
* Allow reasoning pipeline to manipulate state
---------
Co-authored-by: albert <albert@cinnamon.is>
Co-authored-by: Duc Nguyen (john) <trungduc1992@gmail.com>
* Add Adobe reader as the multimodal loader
* Allow FullQAPipeline to reasoning on figures
* fix: move the adobe import to avoid ImportError, notify users whenever they run the AdobeReader
---------
Co-authored-by: cin-albert <albert@cinnamon.is>
* remove cohere as default
* refractor dependencies
* use llama-index pdf reader as default (pypdf)
* fix some lazy docstring
* update install scripts
* minor fix
* serve local model in a different process from the app
---------
Co-authored-by: albert <albert@cinnamon.is>
Co-authored-by: trducng <trungduc1992@gmail.com>
* Create user management page
* Remove old user creating UI
* Add username validation; admin user auto-creation
* Provide docs on user management
* Bump version
1. Introduce the concept of "collection_name" to docstore and vector store. Each collection can be viewed similarly to a table in a SQL database. It allows better organizing information within this data source.
2. Move the `Index` and `Source` tables from the application scope into the index scope. For each new index created by user, these tables should increase accordingly. So it depends on the index, rather than the app.
3. Make each index responsible for the UI components in the app.
4. Construct the File UI page.
* feat: Add installers for linux, windows, and macos
* docs: Update README
* pre-commit fix styles
* Update installers and README
* Remove env vars check and fix paths
* Update installers:
* Remove start.py and move install and launch part back to .sh/.bat
* Add conda deactivate
* Make messages more informative
* Improve kotaemon based on insights from projects (#147)
- Include static files in the package.
- More reliable information panel. Faster & not breaking randomly.
- Add directory upload.
- Enable zip file to upload.
- Allow setting endpoint for the OCR reader using environment variable.
* feat: Add installers for linux, windows, and macos
* docs: Update README
* pre-commit fix styles
* Update installers and README
* Remove env vars check and fix paths
* Update installers:
* Remove start.py and move install and launch part back to .sh/.bat
* Add conda deactivate
* Make messages more informative
* Make macOS installer runable and improve Windows, Linux installers
* Minor fix macos commands
* installation should pause before exit
* Update Windows installer: add a new label to exit function with error
* put install_dir to .gitignore
* chore: Add comments to clarify the 'end' labels
---------
Co-authored-by: Duc Nguyen (john) <trungduc1992@gmail.com>
Co-authored-by: ian <ian@cinnamon.is>
- Include static files in the package.
- More reliable information panel. Faster & not breaking randomly.
- Add directory upload.
- Enable zip file to upload.
- Allow setting endpoint for the OCR reader using environment variable.
