About MYNC
Built for the conversations cloud assistants can't safely hold. For the relationships you'd rather not put in a server you don't control.
The most valuable conversations you have don't happen in your inbox. They happen in iMessage. In voice notes founders send at 11pm. In back-channels where deal flow lives, and in DMs where the real opinions get traded.
For a decade, "AI assistants" have asked you to upload all of it to their servers — and trust that they'll behave. We don't think that's the right trade. Your most private corpus deserves a tool that runs on your own machine.
The pitch isn't "another assistant." The pitch is: your relationships, structured, queryable, available to you wherever you ask — through the AI you already use, against memory you actually control.
And the second pitch: privacy is the brand. Every architectural decision — the encryption substrate, the embedding layer, the agent harness, the way captures sync from iPhone — is made so we can keep that promise structurally, not aspirationally.
The principles every spec, every ADR, every line of code derives from.
Apple Silicon is fast. Storage is cheap. Modern embedding models fit on a laptop. The reasons to move private data to a server have evaporated — what's left is convenience.
We pick the inconvenient path on purpose, because the convenient path makes you the product.
"We don't read your data" is meaningless if the architecture allows it. Every guarantee MYNC makes is enforced by a specific component — encryption, key custody, tier discipline, signed updates.
If we can't enforce it in code, we don't promise it.
MYNC builds a private knowledge graph of your network — and you can edit it. Mark people as forgotten. Override attributes the model got wrong. Exclude noise.
The corpus is yours. The map is yours. The answers are yours. There's no cloud copy to delete.
You already pay for Claude. MYNC plugs in through MCP — your assistant gains a private knowledge graph; we don't ask you to switch tools.
If you'd rather route through our hosted inference, switch tiers. If you'd rather stay fully local, that's the default. You decide.
A frank read on what's live and what's coming.
iMessage ingestion, on-device embedding, the four-layer knowledge graph, local Gemma inference, MCP server for Claude Desktop. Notarized, signed, distributed via direct download.
MYNC-hosted inference for larger models. Same slice-of-what's-needed pattern. Switchable to Local anytime. Includes the iPhone companion.
Query the map from iPhone — routed through the relay to your Mac. Offline captures sync when you reconnect. No corpus replicated to phone.
Source contributors arrive one at a time, each with a privacy posture documented before it ships. The graph grows; the boundary principle holds.