Anoneurx Lab · Self-Teaching AI Research

ASTRA

A self-learning AI that improves through its own experience.

ASTRA is an Anoneurx research project exploring models that learn continuously — recording, reflecting and improving from real use instead of standing still after training. Currently a 20B parameter research system, architected for a 1T long-term target. In development, not production.

Read the research Status

20B parameters1T architectural targetResearch preview

PERCEIVE

RECALL

PLAN

ACT

A · S · T · R · A — autonomous self-teaching research agent

Explore ASTRA

Every part on its own page

Each section of this research project lives as its own page — read them in order or jump to what matters most.

What is ASTRA

The project, the goal and an honest note on the current state.

The learning loop is the whole idea

Perceive, reason, act, record, reflect, improve — the loop never stops at inference.

Built at 20B, aimed at 1T

Where the project is now and where its architecture is designed to go.

A tree with a looping core

Reference design: interface, core, memory and tools around a reasoning loop.

Learns from usage, not just datasets

Record, judge, update, consolidate — the four moments behind self-learning.

The memory pipeline makes recall possible

Episodic, semantic and procedural memory over the raw experience stream.

The open problems that make ASTRA real

Eight active research modules in the Anoneurx Lab — none solved yet.

Current state of ASTRA

Transparent status, no smoothing. What is true today is reported as-is.

A running record, the way research should be

Each entry states what exists and what does not — no release theatrics.

A plan that respects reality

Phases, not promises. Each gate is passed on measured results.

The rules the project won't break

Five principles that constrain every technical decision.

Born in the Anoneurx Lab

The research lab behind the open problems, people and projects.

Help build something honest

Researchers, engineers and critical reviewers — there is a seat here.

quick index

The short route

Start with what the project is, skim the status, then jump to the research problems.

01 · What is 08 · Status 07 · Research

20B parameters today · architected for a 1T target · in development, not production.

The long game

AI that never stops improving — without pretending to be more than it is.

This project will change as the work does. Progress here is written in measured milestones, roadblocks included. Follow the honest record.

Development log Up next · research modules