Levh
Local-first memory for AI agents and workflows
10 followers
Local-first memory for AI agents and workflows
10 followers
LEVH is a local-first memory layer for AI agents and human. It captures context from conversations, projects, files, email, calendars, and MCP tools; builds an inspectable knowledge graph; surfaces trust and conflict signals; and lets stale details fade—without cloud accounts or telemetry.
Finally something that doesn't try to sync my entire brain to a cloud. Tried connecting it to a couple MCP tools and the knowledge graph actually showed me where two notes contradicted each other, which was kind of wild to see laid out.
@ncutigatwa Thank you, Ncuti! Keeping everything local-first was the whole point, so it means a lot that it landed that way. The contradiction surfacing in the knowledge graph is our favorite part too, and it's exactly what we hoped MCP connections would make visible. Hope it keeps running smoothly for you - and if you ever hit a rough edge, an issue on our repo is perfect (or a PR, if you already have a fix in mind). If you feel like helping us push it further, contributions are more than welcome, and sharing it with others would honestly make our day.
honestly the trust and conflict signals idea sounds really useful, but right now it seems like there's no way to actually audit or export the raw graph data. like a simple "export to json or sqlite" button would make it way easier to trust the system and back stuff up, especially since you guys are going local-first. would love to see that added.
@eugene_carllyle Thanks so much for the feedback — really appreciated it. I read your comment and wanted to actually ship the fix before replying. I've added a full export now: JSON, raw SQLite, and a PDF audit report, covering the entity graph, trust scores, and conflict candidates (not just raw memories), so you can audit and back up everything. Tested it end to end and it should work smoothly.
If you'd like to support us, a star on the repo and a share would mean a lot 🙏
Thanks so much to everyone checking out LEVH today! 🙏 Building this has been a labor of love — if you try the quick start, I'd genuinely love to hear what breaks, what surprises you, or what memory problem you wish it solved. Every comment and upvote means a lot on launch day!