A knowledge base that thinks, learns, and evolves every night. Powered by 11 autonomous research engines.
A closed-loop system inspired by ActiveWiki and Andrej Karpathy's autoresearch pattern.
Loading...
Each AI module saves its interactions and learnings. Readable by Hermes for cross-component analysis.
The architecture behind this system is available as an open-source Python framework.
ActiveWiki — github.com/drakkB/activewikiThe Living Wiki is not a static documentation page. It is a self-evolving knowledge base that accumulates findings from 11 autonomous research engines running every night between 1:30 AM and 6:30 AM.
Each engine explores a different dimension: Darwin evolves strategy parameters through mutation and selection. Meta-Harness optimizes Darwin itself — the optimizer optimizes the optimizer. Chimera evolves 1,221 pattern detection thresholds. The Strategic Layer generates testable hypotheses from accumulated knowledge. The newest engine evolves each AI's own system prompt based on its prediction accuracy.
Every discovery — a better RSI threshold, a failed pattern, a winning parameter combination — is written to the Living Wiki as a lesson. These lessons are then injected by PromptForge into every AI call across 14 modules: the Collaborative arena (6 AIs voting on trades), Oracle consultations (9 AIs answering questions), prediction markets, news analysis, the Strategy Genie mentor, and more.
Component Memory adds persistence. Each module remembers its past interactions, creating a feedback loop: the AI acts, remembers the result, and adjusts. Combined with the Wiki's nightly discoveries, this creates a system where every component gets smarter over time — not just from its own experience, but from the collective intelligence of all 11 research engines.
This closed-loop architecture — accumulate, hypothesize, test, learn, repeat — is inspired by Andrej Karpathy's autoresearch pattern and implemented using ActiveWiki, our open-source framework for self-evolving knowledge systems.