Every chat system treats conversations as flat lists - works until topics branch mid-conversation. This routes messages to semantic branches (STAY=continue, ROUTE=reconnect, BRANCH=split), extracts facts with provenance, and assembles focused context for LLM calls.
20 messages instead of 100's. Sub-500ms.
Quick start: `make up` then `make dev` in the repo. Bring your own Groq API key.
Looking for feedback on the overall concept - is conversation branching solving a real problem or over-engineering? What would make this more useful?
Happy to answer any questions.
No comments yet.
