Questions, answered straight
You installed it. Then what?
I installed it — then what actually happens?
After setup (your endpoint + key in the panel), the engine starts on your machine and the extension wires it to your assistant automatically over MCP — no manual configuration. From then on you just work: your assistant gains two abilities, remembering what matters as it happens and recalling it in later sessions. The panel shows your memory growing and the token savings meter running.
How do Copilot or Claude Code know to use it?
Both discover MCP tools automatically — juvina_recall and juvina_encode appear in your assistant's toolbox the moment you reload the editor. The assistant then calls them when the conversation makes memory relevant: recalling before answering about past decisions, encoding when something worth keeping happens.
My assistant doesn't seem to use its memory much. Is that normal?
Sometimes, yes — assistants are new to having real memory, and left uninstructed they can favour what's already in the conversation over what's in the store. The fix is one paragraph of standing instructions telling your assistant when to recall and when to encode — our GitHub repos ship ready-made versions (a Copilot skill, and a CLAUDE.md snippet for Claude Code). And if it still feels off, email us — we genuinely want to hear it: hello@juvina.ai. Early users get the builders themselves, and fast.
What leaves my machine?
Your memories are stored entirely on your machine, in a local database. To build memory, the engine sends the content it memorises — conversation turns and recall queries — to the model endpoint YOU configure, on your own key. Point it at your company's Azure tenant and nothing crosses your own walls. Nothing is sent to us: no telemetry, no phone-home.
What does it cost?
Free forever: your assistant keeps a rolling 7-day memory at no charge, no card. A subscription (£19/month) makes memory permanent — everything your free months stored is still there and comes back in full the day you subscribe. Nothing is ever deleted by the free tier; it simply narrows the window it shows.
Which assistants work with it?
GitHub Copilot (agent mode) and Claude Code today, plus any MCP-capable client via our juvina-mcp connector. Windows-first by design; your model endpoint just needs to be OpenAI-compatible.
How is this different from bigger context windows?
Context stuffing re-sends your history on every call — expensive, attention-thinning, and it still dies at session end. Juvina keeps memory outside the window and recalls compact constructed answers: a few hundred tokens however old the project gets. Curious what repetition costs you today? Run our free counter: npx github:1picassoai/juvina-token-bill
Something's broken, or a question isn't here.
Tell us: hello@juvina.ai. We're a small team shipping fast, we answer quickly, and early-user reports genuinely shape the product — several of this week's fixes came straight from people poking the demo.
Free forever — 7-day rolling memory. No card. Works over MCP.
Get Juvina Light — VS Code Marketplace →Or ask us anything: hello@juvina.ai