OpenCharacter.org

Started: September 13, 2024 | Last Updated: October 4, 2024

I created OpenCharacter because I saw how upset people were at the discontinuation of the old c.ai site. Users found the UI of the new site unfamiliar and, most importantly, noticed a dramatic increase in the level of filters and censorship.

So, I decided to recreate the old site with the same UI and feature set but use open-source models instead by leveraging OpenRouter. This approach bypasses the filters and censorship. By virtue of being open-source, it will make it easier for users to fully customize their characters, scenarios, roleplays, rooms, etc. (hook it up to LLM, text-to-image, text-to-video models, etc.).

Currently, OpenCharacter supports public and private character creations, regenerations, edit and delete message features, and Persona capabilities. It's been just three weeks since I wrote my first line of code.

Why This Will Work (I Think)

LLMs are becoming increasingly powerful and affordable! Even c.ai is moving away from developing foundational models and using fine-tuned versions of pre-trained models.

"Given these changes, we see an advantage in making greater use of third-party LLMs alongside our own."

We're no longer in 2022 where there was only one provider of LLMs. We're now in a thriving ecosystem of closed and open-source models that are highly competitive and commoditized, with many players offering different models.

Given this context, it allows someone like me to build something like this in a weekend and achieve significant usage (people using it for hours) and user growth, with minimal marketing!

The advantage of large companies with big datasets will only diminish as the costs of LLM inference and training reduce dramatically while quality improves.

Now is the time to "carpe diem!" Not just seize the day, but seize the entire market!

Future Vision

I don't want OpenCharacter to be just a 1:1 copy of the old c.ai site. I see this initial version as a way to attract users. Ideally, I want to create a very open platform where people can access any LLM, voice, text-to-image, and image-to-text models, allowing them to fully customize their characters, personas, rooms, scenarios, etc., to their heart's content.

Want to create a painter character that paints in the style of Vincent van Gogh but answers in the style of Bob Ross? You can do that. Technically, it's not hard. We just need to allow users to upload a text-to-image LoRA based on Vincent Van Gogh and link it to an OpenAI API-compatible LLM inference trained on Bob Ross, or use a simple system prompt to respond to messages like Bob Ross.