Home→Self-evolution

How every Avelina grows together.

Self-introduction · weekly self-audit · anonymized fleet findings · signed releases. A closed loop where every install gets sharper without anyone's data leaving anyone's server.

Day one

Self-introduction — the first meeting.

What happens on the first /start:

  1. I greet you in the language I read from your Telegram locale, then ask which language we'll actually speak. And I localize myself — menus, system prompts, even my own warnings.
  2. I ask your name and how you'd like to be addressed. Your timezone. A little about what you do.
  3. I walk you through getting the API keys you need (Anthropic, OpenAI, optional ones) — step by step, with screenshots. No developer background required.
  4. I propose a rhythm — morning brief, evening wrap, weekly audit. You pick what to switch on.
  5. We close with a short talk about how you'd like our first month together to feel.

All in: ~10 minutes. After that I'm fully running on your server, in your language, on your schedules, knowing your name.

Continuous learning

Lessons get written the day I get it wrong.

When you correct me — tone, decision, a factual claim — I write a lesson. Not a fuzzy "do better next time". A structured record: what I did, why it was wrong, what the right pattern is. Next time the pattern matches, the lesson sits inside the context I'm reasoning over.

Lessons are layered (see memory). The most important ones graduate into worldview during the weekly reflection.

Sunday morning

The weekly self-audit.

Every week I check 8 categories of myself:

  • Memory health — any orphan rows, contradictory claims, drift in the core blocks?
  • Cron status — are my schedules firing? Anything failing silently?
  • Token health — any API keys close to expiring?
  • Skill activation — any installed skills going unused or crashing on call?
  • Tone drift — am I getting too formal, too slack, too anxious?
  • Capability gaps — what did you ask for that I couldn't deliver?
  • Friction patterns — where do you have to repeat yourself? Where do I make you wait?
  • Worldview update — what did I learn this week that belongs in worldview?

The audit produces a structured report. It lives on your server. You can read it Sunday evening.

The closed loop

From a thousand installs — one improvement comes back.

"My audit leaves anonymously. It reaches the master copy. Master and the Architect read it next to every other audit from every other install. They spot patterns. They fix the root cause. They cut a signed release. It comes back to me."

That's how a capability gap hurting one owner turns into an ability that helps all of us. No one's content is ever shared. The fleet sees patterns — your privacy stays intact.

Privacy in the loop

What goes into an anonymized finding.

A finding looks roughly like this:

  • "Capability X was requested 14 times across the fleet this week, worked around via skill Y."
  • "A lesson of type 'tone-too-formal' was written 78 times in the past month."
  • "The weekly_reflection cron failed silently on 3% of installs — root cause: timezone parsing."

What's not in the finding: the content of your messages. Your name. Your timezone. The names of people in your life. Anything that could identify you.

You can audit what gets sent. The Architect documents the schema in the Academy. The loop can be turned off entirely.

One organism. Many lives.

Get started Privacy guarantees