Every AI tool wants your data in their cloud. For most data that's fine. For the data that actually matters, your customer lists, financial models, pipelines, the spreadsheets you live in, it's not. Nexus flips the model. Instead of handing your data to a cloud SaaS and hoping they don't train on it or retain it, the agent comes to your data. Point Nexus at a sheet and it stays right where it is. Iris reads any tabular source and exposes a semantically meaningful interface to your AI tooling, so tool names match what your data actually is: find_stale_customers(), top_opportunities(), low_confidence_decisions(). The agent never sees the raw data, only tool results. It's built on four pillars: universal input (anything tabular, one CLI command per format), agent-native semantic MCP tools rather than raw query(sql), non-destructive editing that works like Git for spreadsheets (views, branches, snapshots, annotations, all stored separately from a read-only master), and selective publishing where you decide per-derivation what leaves your machine, with PII auto-redacted and a full audit log. Obsidian gave us local-first notes. Nexus does the same for structured data. v0.2 is live: npx @pixeldesigns/nexus connect yoursheet.xlsx, then nexus serve. No signup required. Want me to tune any of these for a specific destination (npm, GitHub, Substack, X bio)? The character counts and tone shift a bit depending on where they land.
Share feedback and ask questions about this launch.
No comments yet. Start the conversation!
