danny-avila / LibreChat

Enhanced ChatGPT Clone: Features OpenAI, Assistants API, Azure, Groq, GPT-4 Vision, Mistral, Bing, Anthropic, OpenRouter, Vertex AI, Gemini, AI model switching, message search, langchain, DALL-E-3, ChatGPT Plugins, OpenAI Functions, Secure Multi-User System, Presets, completely open-source for self-hosting. More features in development
https://librechat.ai/
MIT License
15.28k stars 2.54k forks source link

Support custom Bing AI #98

Closed HyunggyuJang closed 1 year ago

HyunggyuJang commented 1 year ago

With https://github.com/waylaidwanderer/node-chatgpt-api/pull/199, the customized system message is back; like custom GPT, we should support custom bing AI.

danny-avila commented 1 year ago

Will be part of using the latest node-chatgpt-api and the features i will bring to custom params. I think we can accelerate use of sydney if we get the library working for bing/sydney first through the alias

danny-avila commented 1 year ago

sydney/bing is working thanks to you, I will get on this with my update immediately following the search update

HyunggyuJang commented 1 year ago

sydney/bing is working thanks to you, I will get on this with my update immediately following the search update

Looking forward to! I myself also will look into it on my free time! If you are working on a public branch for this feature, might be able contribute.

danny-avila commented 1 year ago

Not yet started as I’m finalizing searching messages, which will be a big update. It would be really helpful to have bing/Sydney “ask” routes consolidated and cleaned up and that can be right off main. There’s a good amount of code they share but any divergence could be delegated to imported functions

HyunggyuJang commented 1 year ago

Not yet started as I’m finalizing searching messages, which will be a big update. It would be really helpful to have bing/Sydney “ask” routes consolidated and cleaned up and that can be right off main. There’s a good amount of code they share but any divergence could be delegated to imported functions

So, you mean we can work on the bing/Sydney routes decomposition from the other open AI engines (as those are inherently having different way to interact), right? Okay, while you are working on the search feature (very sweet feature!), I'll try to understand and how we design it.

danny-avila commented 1 year ago

So, you mean we can work on the bing/Sydney routes decomposition from the other open AI engines (as those are inherently having different way to interact), right? Okay, while you are working on the search feature (very sweet feature!), I'll try to understand and how we design it.

On the backend specifically, they are split into two different endpoints. This was mainly done for convenience as I figured out how the jailbreak work. I think the two are similar enough to consolidate into one endpoint, where they would share most of the same logic, and where it differs can go into separate handler functions in the same directory.

Having this cleaned up would make it easier to apply custom settings for bing as it can also be in the same endpoint. I plan to use similar language/logic on the front end as with the current customgpt system but we can focus on backend to start,

edit: and by custom bing I mean Sydney’s system message i guess. If you think they should be separate and all custom logic goes to Sydney, I’m fine with that too

HyunggyuJang commented 1 year ago

So, you mean we can work on the bing/Sydney routes decomposition from the other open AI engines (as those are inherently having different way to interact), right? Okay, while you are working on the search feature (very sweet feature!), I'll try to understand and how we design it.

On the backend specifically, they are split into two different endpoints. This was mainly done for convenience as I figured out how the jailbreak work. I think the two are similar enough to consolidate into one endpoint, where they would share most of the same logic, and where it differs can go into separate handler functions in the same directory.

Having this cleaned up would make it easier to apply custom settings for bing as it can also be in the same endpoint. I plan to use similar language/logic on the front end as with the current customgpt system but we can focus on backend to start,

edit: and by custom bing I mean Sydney’s system message i guess. If you think they should be separate and all custom logic goes to Sydney, I’m fine with that too

Got it! Thanks for your clarification! Now I get the picture. So rather than decomposing, but make use of existing logic if possible. Okay, I’ll try to look into it.

Erisfiregamer1 commented 1 year ago

You're welcome, by the way.

danny-avila commented 1 year ago

You're welcome, by the way.

Yes, thank you for implementing this in the node library and for finding your way here! I’m working on this now

danny-avila commented 1 year ago

Finished in #142

HyunggyuJang commented 1 year ago

This is epic! Thanks a lot.