Hi PH 👋 — solo builder here. I got tired of my coding assistant turning into a stranger every session: re-explaining my own architecture daily, and paying tokens for the privilege. So I spent months building Juvina — a memory engine that runs entirely on your machine (local Postgres, your own model endpoint and key) and connects to Copilot/Claude Code over MCP. Sessions encode decisions as they happen; any later session recalls a compact constructed answer instead of replayed history — so the token bill goes down as the project ages, not up. There's a no-install demo at juvina.ai: ask about a codebase's design decisions, destroy the context, ask again — and watch what survives. Free forever (7-day rolling memory); a small subscription makes memory permanent. Brutally honest answers to any question — architecture, privacy, what it can't do yet. Ask away.

Juvina

Persistent, local AI memory for Copilot & Claude Code

Persistent, local AI memory for Copilot & Claude Code