[Bug] Can't get Wingman to work on Windows

[djsudduth]

Thank you @dvorka for the recent updates to mindforger 2.0!! Super exciting to see.

However, I can't seem to get Wingman to work for me. I'm using Windows 10 right now with my OpenAI key. I was able to confirm the key works with other tools. But, every time I run Wingman I get
ERROR: Error: request to OpenAI Wingman provider failed due a network error - Error creating SSL context ()

Any ideas? I tried the new key and model with Logseq and it worked fine - no network error.

Also, any way to pick the LLM? I typically use gpt-4-turbo.

[dvorka]

@djsudduth apologies, I'm still releasing and adding distributions for various operating systems. Can you please try (better) Windows distribution which is based on Qt 5.15.2?

• windows-installer-mindforger-2.0.0-5a430819-qt5152.exe

I just uploaded ^ to the MindForger 2.0.0 release page. Check that you Preferences look like in the screenshot below and OpenAI is chosen as LLM provider:

Also, any way to pick the LLM? I typically use gpt-4-turbo.

Unfortunately it is not possible to choose the model in MindForger 2.0.0, however, I will add it in the next release - it will be simple enhancement.

[djsudduth]

@dvorka - installed the update and confirmed the configuration. I now have a new error:

ERROR: Error: request to OpenAI Wingman provider failed due a network error - TLS initialization failed

[dvorka]

@djsudduth let me try to repro that on a Windows machine different from the one I use for MindForger development. Thank you for reporting the 🐞 !

[djsudduth]

@dvorka - you probably knew this - but in case it helps: https://stackoverflow.com/questions/53805704/tls-initialization-failed-on-get-request

[dvorka]

@djsudduth thank you! 🐞 is reproducible on non-development Windows machine. I didn't fix it there yet, but the most probable root cause is missing OpenSSL and/or version mismatch. I will need to find out how to include missing libraries in the installer.

---

Problem:

• Qt for Win does not have OpenSSL included.

Solution:

• include OpenSSL DLLs to MindForger .exe path - there are two options:
  • next to mindforger.exe binary
  • in Windows/System32 directory
• DLLs to be included:
  • qt < 5.12.4 (DLLs can be found in bin/ directory of OpenSSL folder)
    • libeay32.dll
    • ssleay32.dll
  • qt >= 5.12.4 (*)
    • libssl-1_1.dll
    • libcrypto-1_1.dll
    • capi.dll ... dependency of above libs (they need it)
    • dasync.dll ... dependency of above libs (they need it)
• why?
  • it is known Qt 5.15.2 issue that it requires OpenSSL 1.1.1
    • https://wiki.qt.io/Qt_5.15.2_Known_Issues#:~:text=for%20Automation%205.15-,OpenSSL,libraries%20used%20by%20Qt%20Network
• where to get DLLS?
  • get OpenSSL from Shining Light Productions
    • https://slproweb.com/products/Win32OpenSSL.html
    • choose Win64 OpenSSL v1.1.1w Light (EXE)
    • install it w/ .dll extraction to bin/ directory (not System32/)
  • copy all DLLs from above (*) to bin/ next to mindforger.exe

Installer:

• mindforger-setup.iss:

[Files]
Source: "C:\Path\To\OpenSSL\bin\*.dll"; 
DestDir: "{app}\bin";

[dvorka]

@djsudduth I think that I solved problem with OpenSSL on Windows - please try MindForger 2.0.1 (either link below or from the Releases):

• https://github.com/dvorka/mindforger/releases/download/2.0.0/windows-installer-mindforger-2.0.1-20365ac5-qt5152.exe

As a small thank you for your help, I've added the ability to set your favorite OpenAI model using environment variable 😉

# Windows
set MINDFORGER_OPENAI_LLM_MODEL=gpt-4

# Linux/macOS
export MINDFORGER_OPENAI_LLM_MODEL=gpt-4

Please give it a try!

[djsudduth]

@dvorka Fantastic!!! It works very well! I tried some variations on notes and no issues at all. Thank you for the effort to fix it and adding the LLM option!

Side question: were there any enhancements to the ML 'thinking' model for associations?

[dvorka]

@djsudduth 😌

were there any enhancements to the ML 'thinking' model for associations?

I did not improve associations in MindForger - it still uses a pretty simple algorithm that is not bad, but it also does not work sufficiently well.

General-purpose LLM integration (like OpenAI) is meant for use cases that assist in note-taking. Associations (semantic similarity search) and "talk to your notes" will be about a RAG-like solution.

I would like to improve associations and have some fun. My plan is to use word embeddings (I'm looking for a fast model to calculate them) and a quick similarity search like https://github.com/facebookresearch/faiss. The most challenging part will be implementing a fast, local (notes will not be sent anywhere), CPU-only solution that will use a reasonable amount of resources.

Anyway, I'm looking forward to it!

[djsudduth]

Yeah, faiss makes good sense to try. There are plenty of RAG solutions to choose from for vector store and retrieval. Looking forward to what you might come up with. Thanks!