Claude Code loses everything between sessions.

A long session builds real shared understanding — the plan, the constraints, the decisions and their reasons. Then the context fills up, or you /clear, or you close the terminal. Next session: a stranger again.

Why it happens

Claude Code works inside a context window. Long conversations get compacted or cleared, and a new session starts empty by design. CLAUDE.md files help with standing instructions, but they're static notes you maintain by hand — not a memory of what actually happened and what was decided.

What it costs you

Re-explaining is the visible cost. The invisible one is tokens: a long-running chat re-sends its accumulated history again and again, and every fresh session pays to rebuild understanding the last one already had.

The fix

A memory that lives outside the context window. Juvina Light runs a local memory engine on your machine and connects to Claude Code over MCP: sessions encode what matters as it happens, and any later session — tomorrow, next month — recalls it as a compact constructed answer, not a replayed transcript. Everything stays on your machine.