Built for one person at a time.

Mayva started as a private tool. One developer's morning ritual: read recovery, scan the calendar, dispatch a few small actions, decide what the day actually is. Repeated for months. Then automated.

We built Mayva because the existing options were either too cloudy (everything stored somewhere else, charged per seat, locked behind APIs) or too dumb (a cron job that emails you a CSV). What we wanted was an agent — something that knew our day — but one whose memory we could grep and whose prompts we could read.

The principles

1. Local by default. Your data lives on your machine. The LLM call is the only network hop. Sync, if you want it, is encrypted and opt-in.
2. Readable everywhere. Profile is markdown. Goals are markdown. Episodic events are SQLite rows you can sqlite3 into. Every layer is grep-able.
3. One brain. Workers don't have logic of their own. The dashboard doesn't have a separate model. Adding a capability is one place.
4. Quiet, not chatty. The agent earns trust by being right and quiet, not by being talkative. Suggestions are priority-sorted, dismissible, and don't expect a reply.
5. Hackable. TypeScript strict, Zod at every boundary, MIT-licensed. The prompts are strings in files. Change them.

The company

Mayva is built by CorporateThings, a small studio building agent-first software for technical professionals.

The hosted version of Mayva funds the development of the open-source version. There is nothing locked behind the subscription you can't grab from the repo.

Honest disclaimers

• This is not medical advice. Mayva reads your wearable data and adjusts its tone. It doesn't diagnose anything.
• Voice is macOS-only today. Depends on sox, say, afplay. Everything else works on Linux too.