01 · Project
A model that keeps learning after training
Most models are frozen the moment training ends. ASTRA is an attempt to close the gap between 'trained' and 'finished' — a system built to keep getting better from real experience.
What ASTRA is
A research AI with a continual self-improvement loop. It perceives a task, reasons about it, acts, records the outcome, reflects on its own performance and updates itself — without waiting for a human to retrain it.
What ASTRA is not
Not a chatbot, not a hosted product, and not a magic AGI claim. No inflated benchmarks, no promised superintelligence. It is a slow, careful engineering attempt at one specific property: learning from experience.
Current state
A 20 billion parameter research system in the Anoneurx Lab. The learning-loop simulation is the active phase; memory, self-evaluation and task generation are being built on top. Everything below is honest about what exists and what does not.