Added o3-mini Support + Update to llama 3.3

DuckDuckGo Chat has recently added a new model to its website, o3-mini. Additionally, they no longer are using llama3.1 and have updated their model to use llama3.3, despite their 3.1 endpoint still being up (Thank you @realgorlix for pointing this out!). This mini release is to add support for these new models.

Docker Image has been updated, you can find it here on Docker Hub.

Speech API Endpoints + Reintroduced Bug Fix

Notes

1. Added audio generation endpoints, which is based on the OpenAI Speech API . It uses TikTok's Text-To-Speech to provide the voices (there's a ton to use). This is useful for those who run their own Web UI's for LLMs and would like to have their messages read to them. Or perhaps you are a developer who is making/prototyping a voice-based assistant but you do not want to pay for speech generation APIs during development.

The speech API can either be used alongside the /completions endpoint (to "hear" your LLM's responses) or it can be used independently so you can generate speech based on your own input text, again mimicking how OpenAI does it for compatibility. Refer to the website to see the sample API calls.

• /v1/audio/speech/
• /v1/audio/speech/voices

If you are interested in what voices you can use you can take a look here

Here is a demo of it in action

Screen.Recording.2024-11-20.at.2.37.22.AM.mov

These endpoints are completely optional are not functional without setting a session_id as outlined in the Wiki

3. The partial-context bug that was supposed to ship with Version 1.0.0, is now shipped to this current version. I had made the mistake of not applying the bug fix to the branch I was working on at the time. But all is well now.

4. A Wiki has been made for setup instructions, so it declutters the main README file.

5. The website has been updated for this release, check it out

Feel free to pull the latest updated Docker image from callbacked/keyless:latest

Remote Docker Image + Bug Fix + Responsiveness Improvement

Notes

• Opened a repository on Docker Hub for easier deployment/updates in one command, so users don't have to build images manually while keeping up with updates see more information on how to deploy here under "Docker". Or refer to the README.
• Fixed a major bug in #8 where context from the AI's responses would not be saved, causing the AI to not remember its own previously sent messages. It would be best to update.
• Additionally in #8, messages are now streamed in true real time, as opposed to simulating chunks of text being sent, which made the responses feel slower. Now the responses are as fast as the requests get handled at DuckDuckGo without constraint.