Architecture

My Alicia is a thin core surrounded by composable skills, with three loops running over a shared memory substrate.

Alicia's architecture is inspired by many generous thinkers — among them Garry Tan and Andrej Karpathy. Thank you.

The shape of the system

The three relationship loops of myalicia Listen, Notice, and Know — three depths of attention that compound over time, sharing a memory substrate, reachable through Telegram and Claude Code surfaces. Memory vault · notes · archetype Listen seconds · be present Notice minutes–hours · catch patterns Know days–weeks · come to know conversation event-triggered synthesis scheduled reflection Telegram conversational Claude Code technical an AI teammate that grows into the shape of the person it serves

Listen, Notice, Know — three loops, two surfaces, one memory.

Thin harness, fat skills

A small, opinionated core handles orchestration and routing, surrounded by many composable skill modules that each do one thing well. The core is intentionally small because the skills are the contribution surface — people should be able to add a skill without understanding the whole system. 

myalicia/
├── core/           # the thin harness — loops, scheduler, routing, memory
├── skills/         # the fat skills — ~80 modules, each composable
├── surfaces/       # telegram, claude_code, cli adapters
└── config.py       # the single source of truth for personalization
Anatomy of a skill A skill module has a trigger, reads from the memory substrate, processes, writes effects back, and lives at one of the three loops. TRIGGER message · event · schedule SKILL MODULE read vault memory archetype think analyze synthesize decide write response note action lives at one loop Listen · Notice · Know SUBSTRATE vault · notes memory files archetype EFFECTS user reply vault note tool call OUTPUT surface · memory · trigger A skill is a small, composable unit of perception. Many skills make a teammate.

What's inside a single skill module — the contribution surface for designers and builders.

The three loops

Each loop runs at a different cadence and uses a different model tier. Cheap by default — turn the dial up only where depth matters.

Loop Cadence Model What it does
Listen seconds Haiku conversation, in the moment
Notice minutes–hours Sonnet event-triggered synthesis
Know days–weeks Opus scheduled autonomous reflection

Self-healing, self-extending, self-aware

Three traits we treat as teammate qualities, not internal plumbing:

How My Alicia grows USE at the top branches into three parallel growth mechanisms — self-healing, self-extending, self-aware — which converge into "every instance grows." USE daily relationship SELF-HEALING trigger errors action log trajectory result fix in next loop SELF-EXTENDING trigger behavior gap action research the gap result new skill authored SELF-AWARE trigger uncertainty action confidence check result model escalation EVERY INSTANCE GROWS sharper with each cycle Three mechanisms, one teammate, getting better with use.

How My Alicia improves with each interaction.

Self-healing

When a skill errors, My Alicia notices, logs the trajectory, and proposes a fix in the next reflection loop. You don't have to babysit her.

Self-extending

My Alicia can author new skills based on observed gaps in her own behavior. Every instance grows.

Self-aware

Every action runs through metacognitive confidence assessment; uncertain actions escalate to deeper models automatically. The system knows what it doesn't know.

Hybrid surfaces

Reach My Alicia through Telegram (conversational), Claude Code (technical), or the CLI (anywhere). The same teammate, different surface for different parts of your life. Surfaces are pluggable — each is a small adapter wrapping the same handle_message pipeline, so contributors can add Discord, iMessage, voice, or anything else.