Hey Product Hunt! 👋🏻 I'm Bhuvan, solo founder of Junê. Junê was born inside another product. Early this year I built August, a multi-agent assistant — a supervisor coordinating specialized agents. Its weakness became the insight: my agents couldn't share context, so any question that needed their combined knowledge was unanswerable. I built a knowledge graph into August to fix it, and that layer worked so well I extracted it as its own product. The bet under Junê is simple: everyone in this category treats memory as an LLM problem — models running at almost every step, with the cost passed on (the hosted clouds run $19–475/month). I treat it as an architecture problem: do the expensive extraction once into a living graph, make every model optional and swappable, and the same accuracy gets 20–25× cheaper. That's also why the free desktop app can be genuinely local-first and private — your files, your graph, and your keys stay on your machine. Two things I'd love you to try: 1. Drop a messy folder into the desktop app — PDFs, spreadsheets, notes — and ask a question that spans them. The answer comes back cited. 2. If you build agents: wire the MCP server into Claude or Cursor and give every agent you run the same graph. And don't take my word on the numbers — pip install june-bench reproduces the head-to-head against a funded competitor in one command. Claims are cheap; I ship the harness that checks mine. I'll be here all day — honest feedback welcome, especially the critical kind.

Junê

One living knowledge graph any agent thinks from

One living knowledge graph any agent thinks from