Install · Docker
Memory for your AI, anywhere Docker runs.
Mac, Linux, Windows — the engine and its database run in containers on your machine. Your memories live in a local volume; model calls go only to the endpoint you configure, on your own key. Nothing is sent to us. No telemetry.
Step 1 — get the kit
mkdir juvina && cd juvina
curl -fsSL https://juvina.ai/docker/juvina-docker.zip -o juvina-docker.zip
unzip juvina-docker.zipStep 2 — three values
cp .env.example .env
# edit .env:
# OPENAI_BASE your OpenAI-compatible endpoint (no trailing slash)
# JUVINA_SETUP_KEY your model API key — used only against YOUR endpoint
# PG_PASSWORD any strong password for the local databaseStep 3 — up
docker compose up -d
curl http://localhost:8080/health # up when this answersConnect your assistant
# Claude Code
claude mcp add juvina --transport http http://localhost:8004/mcp
# Claude Desktop / any stdio MCP client
npx -y github:1picassoai/juvina-mcpTools: juvina_recall · juvina_encode. Semantic Kernel agents: the Juvina.SemanticKernel package (coming to NuGet).
The image: docker.io/picassoai/juvina-light. Memory quality is certified with gpt-oss-120b-class chat models and text-embedding-3-large — smaller models degrade recall and are not supported. Your volumes are yours: deleting them deletes your memories.
Prefer VS Code? Juvina Light on the Marketplace installs the engine with one click on Windows.