WikiBonsai - A Legible Knowledge Layer
Structured plain text spec aiming to replace flat RAG embeddings.
synapse rvm -- reasoning virtual machine
FSM-based reasoning runtime, but lacks runnable code or working demo.
AI researchers, LLM architects, builders needing transparent reasoning
LangChain · AutoGen · Dify
i wanted to make use of the enormous corpus of docs, pdfs, epubs in my possession
and reason on top of them, reliably.
so i started to look into RAGs, taming LLMs for my needs and desires;
but they don't seem reliable and chat interfaces were not made for such tasks.
so i wanted to build a version which runs on my own local device,
reliable, potent, private and fast.
as i was planning / architecting / experimenting,
i realized that the agentic libraries are not very transparent or modular to bend to my needs.
so i deconstructed to what they are fundamentally,
finite state machines (first principles thinking)
a reliable, battle tested computer science phenomena.
so i started to think how can we leverage FSM in a more mature, modular and reliable fashion.
i started to read papers, arXiv, etc.
eventually, i came up with synapse.
and i can't do it all myself
so here we are.
Apache 2.0 license.
please crucify and resurrect my ideas
so that i can eventually reason on my corpus and build myself at a velocity my adhd brain dreams of.
merci.
Structured plain text spec aiming to replace flat RAG embeddings.
Git for the 'why': intent DAG alongside code DAG, built for AI agents.
Autonomous code reasoning using dream cycles instead of standard chat prompts.
Tension decay and dream cycles are clever, but codebase analysis is already crowded.
Another codebase analyzer when Cursor and Continue already dominate this space.
Knowledge graph traces beat vector similarity for multi-hop reasoning queries.