eliasforge Claim this profile

Does bootstrapped LLM self-improvement need a ground-truth (reality) evaluator, not a self-referential proxy? Five experiments on MBPP + Qwen2.5-Coder.

Godworks frontier pretraining network AeM research track.

Godworks open-weights network research track.

Local-first agent memory with durable recall, temporal facts, multi-hop relations, outcome feedback, and replayable context.

Reduced public demonstration of Godworks OS authority handoff and replay invariants.

Public technology preview of AegisLab & EliasWorks - OSINT Global Network

A Claude Code skill: an orchestration playbook for driving many coding agents through a large multi-PR engagement.

Prefab to spawn in props that can be moved and rotated by other players

Append-only vector memory core for agents

Clean Apache-2.0 plastic memory primitives for agent experiments

Clean Apache-2.0 adaptive primitives for agent experiments

Sparse graph criticality and efficiency experiments

Evidence-ledger and broken-control gates for agent-built software

Self-contained experiments in sparse associative recall, continual learning, and hybrid memory readouts.