Juvina Light — memory for Copilot & Claude Code

Free forever. Your AI remembers.

No card, no clock, no lock-out.

• Free — rolling 7-day memory
• £19/month — permanent memory

Nothing is ever deleted by the free tier; upgrade any time and everything the engine quietly kept comes back.

Your AI bill went up. Your work didn't.

Every session, your assistant is sent the same history it was sent yesterday. The same architecture notes. The same decisions. The same explanations you already paid to send once.

You are not paying for new work. You are paying to say the same thing again.

There is no line on the invoice that says so.

And every morning it starts at zero

Two hours explaining your codebase. Tomorrow you do it again.

The code survives — git makes sure of that. The reasoning doesn't. Why you chose that approach. What you tried that failed. The constraint someone found three weeks ago that nobody wrote down.

Your assistant can read every line you own. Ask it why and it has nothing.

See it work before you install anything

juvina.ai/demo — a working .NET codebase you can interrogate right now. No install, no signup, no card.

Ask it:

"Why is the unit price captured on the order line at order time instead of read from the product?"

It answers:

"UnitPriceAtOrder is copied from the product the moment the order is created — so changing a product's price later can never silently rewrite invoice history."

That sentence is in no file. Nobody wrote it down. It was constructed from decisions made in different conversations, on different days.

Ask it what broke on the first smoke test. Ask it whether customers can be deleted. Then try to catch it out — that part is genuinely useful to us.

Your key. Your machine. Your data.

Your memories never leave your machine. There is no Juvina server for them to leave to.

Model calls go straight to your own endpoint, on your own key — you pay your own token bill, under your own provider's terms. Point it at your company's Azure AI Foundry tenant and nothing crosses your own walls.

We hold your email address and a licence record. That is all we could lose.

What does reach your model provider: to build memory, the engine sends the content it memorises — your agent's conversation turns and your recall queries — to the endpoint you configure, for tagging, embedding and answer construction. That happens under your own account, exactly like any other call you make with that key.

Setup — five clicks

1. Install the extension, restart VS Code, open the Juvina panel from the activity bar.
2. Enter your OpenAI-compatible base URL and your own model API key. Stored in VS Code secret storage. Juvina never issues or sees keys.
3. Install engine — one-time download, SHA-256 verified.
4. Start — the engine boots on localhost.
5. Connect Copilot and/or Connect Claude Code — one memory, both assistants.

Your assistant now has juvina_recall and juvina_encode. That's it.

What you should know before you try it

It constructs — and it can construct wrong. Juvina doesn't look up stored sentences; it builds an answer from what it has. That is the whole point, and it is also the honest limit. If it gets something wrong in the demo, we'd rather you found it than didn't.

Windows x64, VS Code 1.90+. Windows-first by design. Want another platform? Say so at juvina.ai — demand decides.

You need an OpenAI-compatible endpoint — OpenAI, Azure OpenAI / AI Foundry, Groq, Mistral, OpenRouter, Gemini's OpenAI-compatible endpoint, or similar. It must serve both chat completions and embeddings. This powers the engine's own tagging and construction. An Anthropic key can't power the engine (different API, no embeddings) — but Claude Code itself works exactly as normal alongside it.

Model notes: gpt-oss-120b is the certified chat model. The embedder (text-embedding-3-large by default) becomes your memory's vector space and can't be changed later without rebuilding.

Free tier and Pro

Free forever, no card. Nothing to request from us — the only key involved is your own.

• Free: rolling 7-day memory. The engine never stops, and nothing is ever deleted — older memories just rest beyond the window.
• Pro, £19/month: permanent memory — and every older memory the engine kept wakes back up. Cancel any time; you fall back to free, never to nothing.

Running a team? hello@juvina.ai

Built in the open, and we answer

Juvina Light is young, and we are close to it. If something breaks or feels rough, tell us — hello@juvina.ai. A real person answers, and every note shapes what ships next.

/engine/Juvina.exe      the memory engine
/engine/Prompts/*       engine prompt assets
/pg/                    embedded PostgreSQL binaries (pgvector included)
/bootstrap.ps1          first-run setup: init data dir, create db, enable pgvector, seed config

Published by DATAVERSE CONSULTING LTD (juvina.ai). Extension: Apache 2.0. The Juvina engine is proprietary and licensed separately — see LICENSE.txt.