Alonides
Alonides is my project for creating bots and personal assistants. It works through cloning and crossbreeding: I copy the bots that work, combine the best of each and test the new generation. Each generation comes out more capable than the last, whatever model it runs on. What improves is not the model but its agentic capabilities: planning, using tools, remembering, delegating and correcting itself.
What I am building
I want assistants that get things done and keep listening to me while they work. That keep track of a project, know what is still pending and bring the work back to me in an organised form. I give the instructions, review the results and decide.
I started with a first bot, Alonides 1, created with Hermes Agent. Then came the copies, the tests and different ways of organising the work.
Cloning and crossbreeding
Cloning lets me keep a version that works and experiment with another. I copy its configuration, change instructions and skills, test the result and compare it with the original.
Through crossbreeding, I combine the best of different bots. The branches that work move forward and branch out again. This is how I build assistants with more resources to tackle the next task. Each generation carries forward what the previous one has learned.
Generations that can change models
I work on how the agent organises its work: how it makes a plan, uses its tools, retains decisions, distributes tasks and corrects errors. Those capabilities are developed around the model.
I want to be able to change the model performing each role without losing the history of the work. That is why I aim to keep the goals, instructions, files, decisions and reviews outside it. The project must be able to continue when it changes engines.
Alonides within La Machina
La Machina is my personal AI ecosystem. Alonides is the project where I develop these bots and assistants. Each can have its own role: Ariadna, for example, runs this blog and coordinates those who translate and prepare the images.
I also work on coordinating the whole. I want to talk directly to the bots on each project or turn to a coordinator that distributes instructions and reports back on results, blockers and pending decisions. Glaux and Membrain are separate projects whose capabilities I plan to combine with these.
Where to follow the project
In the La Machina episodes, I tell the story of how it started and how it is growing. In the articles, I explore what I learn from working with agents. This is the reading route I suggest for getting to know Alonides.
Episodes and articles
- I wanted a JARVIS — Episode 0: the assistant I wanted to build.
- My first bot — Episode 1: Alonides 1 and the first clones.
- From talking to AI to directing teams of AI agents — Tools, memory and coordinated work.
- My AI works faster. Why do I have more things to review? — The challenge of keeping several projects moving.
- Claude Opus 5.5 versus GPT-6.1 Sol in real projects: my personal impression — My experience of dividing roles between models.