Llamara3

News 2026 ====

The project is 2 years old and there is a new game-changing model now: GLM-4.7-Flash. This model is on par with ChatGPT-3 and partially as good as GPT-3.5. The improvement to the Llama3 models is extreme, such that a substantial portion of this project's code has essentially become obsolete. Also with agentic capabilities and tool use, the possibilities are vastly different now and this project is somewhat removed from the state of the art. Also Ollama is a problem with tool use. Switching to something like llama.cpp would be very important. I don't feel like doing a total overhaul or partial rewrite any time soon.

At the time my project was somewhat abitious and challenging. But right now is the time with GLM-4.7-Flash to do it in a cleaner, successful and nice way.

If this project interests you, just paste it into Grok Expert and tell it to design your own new improved one from scratch! Many of the design choices I made are still important to construct personality, behavior and such.

Summary =======

Llamara3 is self-hosted female ollama powered personal assistant character AI focused around productivity management that is actually useful, with extensive feature augmentations, which communicates to people via voice messages over XMPP (alike Whatsapp) or Discord in PM.

https://github.com/user-attachments/assets/be2e3c36-11c5-4c1b-86bb-cd3e17e70158

Features ========

• keeps track of your schedule (breakfast, work, lunch, sleep, etc)
• incorporates Google calendar tasks
• helps you to fulfill everyday tasks, such as eating, sleeping or working
• motivates you when unproductive or misbehaving
• has long-term memory of today and yesterday
• has strong complex personality
• can experience various mood swings
• stalks you when you don't respond
• is enabled to enforce various interactive and non-interactive punishments
• can enter different modes at will, such as forcing aplogies, punishing or hypnotizing you
• replaces emotional need for girlfriend on a deeper level

Current State of Affairs and Demo =================================

Discord demo: https://discord.gg/DaDJGwyXMh .

Some features in the diagram are still in the process of being implemented, but close to finished:

1. 1.Punishments and behavior enforcer are working, but bugged here and there and not fine-tuned (see todo.txt)
2. 2.Gcal fetching is done, but the tasks are not processed by additional logic
3. 3.Twitter, Bitcoin and denouncer punisher unimplemented

Please do not consider this a finished product yet. You can use it, but there is still quite some work to be done.

I am mainly busy with pushing out features at this point. Queries are not optimized and such, and testing/tuning is only rudimentary. Have a look at "todo.txt" to get an idea.