AGL will be usable (one day)
This blog post was released on the eighteenth of August 2026.
New version?
For now, I haven't decided what the best way to version agentLIBRE is. It is in alpha, and I am just trying to see what is really needed and what is an unnecessary complication. It is so early in development that it does not really support thinking.
At the same time, the newly released Qwen 3.8 27B supports reasoning effort. Extremely cool and great—Qwen delivers as always. It is too early to evaluate the performance, but it is definitely an improvement.
The goal
My aim is to make it easy for people to use some of the modern "AI"-enabled tools without dealing with the pain you usually experience when setting everything up. I want to start small and see where it leads me.
ChatGPT
Proprietary models conceal a lot, and that makes it difficult to maintain a consistent understanding of their performance. Nevertheless, sometimes I let Codex implement something the way it sees fit—overcomplicated and with strange fluff words.
My code review flow is improving, and it is becoming easier to manage all the GPTsms. Still, there are some components that I will probably separate into different projects/repositories.
Continuous Integration
I have some scripts that serve CI/CD purposes, and I had to fight with them for some time. The fact that GitHub died yesterday did not help. I use GitHub just for the sake of its compute resources.
Some big FOSS projects use GitHub mirrors even if they use different platforms to host their code. I don't think it is really possible to protect yourself from scraping, so for now I embrace it.
What's next?
I hope to continue the process of integrating Matrix into this website. It would be great if people told me why I am wrong in the comments right here.