Changes

• Now uses Ollama Embeddings instead of FastEmbed

Additions

• Now allows selection of embedding models from Ollama

Fixes

• Removed unused dependency (transformers)

Additions

• Added swagger to flask app for easier testing and doc

Fixes

• Minor bug fix in flask app, unintentionally writing settings to settings.json, causing errors

Fixes

• Now checks if metadata exists before uploading to Neo4j
• Now always reads password as a string

Fixes

• Fixed main.py not using Neo4jvectorstore

Fixes

• Fixed backend not updating connection settings
• Fixed backend not properly adding new documents
• Fixed backend not respecting system arch when starting neo4j

Changes

• Modified to a single working platform-agnostic exe

Additions

• Settings and Prompts exposed to UI

Improvements

• Added date to every session created for better division

Additions

• LLM select and pull options
• Additional metadata for files
• Folder selection

Fixes

• Localized Bootstrap
• Fixed cleanup not properly terminating ollama server

Improvements

• Implemented streaming output (chatgpt-esque)
• Uses chrome as browser, if available

Fixes

Added cl100k_base.tiktoken in tiktoken_cache for fully local and offline running: llama-index, by default, uses this tokenizer

Future fixes

Use bootstrap locally for CSS

First Release of ToK 🎉

Install the attached .exe, run it (ignore the warnings 😜), and done! (provided you have completed all the steps in README.md)

Kindly provide your feedback and let me know any improvements that can be made.

Thanks!