Wrote super quickly, as a fun project, this is a wrapper for GPT 4 (including Vision), Whisper, ElevenLabs (and XVASynth) for extremely flexible AI assistants - in this case made for Star Citizen but can be changed for anything/
Install NodeJS and any required dependencies. Then npm i
and node index.js
Make sure to go in confs/api-keys.json
to change your API keys for ElevenLabs and OpenAPI.
- OpenAI GPT-4
- ElevenLabs
- UNEXCORP (https://twitter.com/CardilloSamuel/status/1727031996818330020)
- Leave the assistant key empty, a new one will automatically be created by the software
- If you go in the personas, you can delete the
threadId
individually, a new one will automatically be generated when relaunched, giving you an entirely clean context to play with - Assistant tools are updated automatically everytime you launch the code so you can have fun with
confs/tools.js
- You can use XVASynth instead of ElevenLabs (saves money I guess) by modifying the personas JSON.
- An HTTP(S) server is running automatically on port
:5446
(you can change on/confs/server.json
) with a working client-server exemple (files in/webfiles
)
You can easily add tools in confs/tools.json
and then the correct function in openai_functioncalls.js
. The assistantKey
in the confs/api-keys.json
is automatically filled.
- https://nodejs.org/en/download
- https://sourceforge.net/projects/sox/
- https://www.java.com/fr/download/manual.jsp
What's already planned :
- Refactoring of the Vision API to be useable in a standalone function
- Adding an interval for uploading files & refactoring of the "temporary assistant" to be able to retrieve existing files (to avoid uplodating/deleting all the time)