Start free

What shipped

Release notes

A plain record of every published version. No roadmap promises here — only what is live.

Beta 0.1

CurrentAugust 2026

Everything built since inception, in one record: the public site, her setup, the skill library with real written instructions, the approval queue, the outbound handoff layer, her own AI endpoint, the scheduled skill runner, the admin panel, and the pitch deck.

The public site

  • Landing page built around the promise — less load, more capacity, time returned.
  • Manifesto, Privacy and Terms pages in the same editorial voice.
  • Brand system anchored on deep plum with rose accents, Playfair Display and Inter.
  • Original line-art illustration set replacing stock photography throughout.
  • Horizontal wordmark with a light-on-dark variant that switches by background, plus the lotus mark as favicon.
  • Waitlist capture hardened: validated email, no duplicates, write-only from the browser.

Accounts and setup

  • Sign in, sign out and an OAuth callback that lands her inside the app, not back on the homepage.
  • Signing out returns her to the homepage.
  • Eight-step setup capturing her household, her organizations, her rhythms, what a good week looks like and what she would do with an hour back.
  • Organization domains accept full URLs, emails or comma-separated lists and are normalized automatically, so school and work mail is matched correctly.
  • Her context is stored per account with row-level isolation — no one else can read it.

My Own — her home surface

  • A personalized home showing what needs her, what is already handled and where to go next.
  • Mobile navigation rebuilt: a fixed bottom tab bar, with account and sign-out moved into a header menu.
  • Live counts on the home card and in the nav when something is waiting on her.

The skill library

  • Twenty skills, each with a full written instruction body, guardrails, required context and an output contract — not prompt one-liners.
  • Doer skills: week ahead, school comms, appointment wrangler, inbox triage, family ops, morning brief, Sunday reset, meal plan, travel brain, pet care, home base, gift closet.
  • Dreamer skills: one hour back, mental load dump, boundary builder, money calm, hard conversation, the thing you want, career momentum, season view.
  • Catalog de-duplicated and re-tiered so free and premium skills read cleanly.

Needs You — the approval queue

  • Every drafted action waits for her: approve, edit, not now, or never do this again.
  • 'Never do this again' writes a standing rule into her boundaries so it does not come back.
  • A 60-second undo on every decision.
  • Nothing leaves without her word — the queue is the only path to an outbound artifact.

The handoff layer — approval means draft, never send

  • Approving an email places it in her email client as a draft, written exactly as she left it. She presses send.
  • Optional Gmail connection held by HER OWN itself, drafts only: one tap puts an approved email in her Gmail drafts. There is no send path anywhere in the code.
  • The rule holds identically whether she approved it in HER OWN or in her own assistant — an assistant with mail access is instructed to draft, never send.
  • Calendar items stay a finished artifact she adds herself: a Google Calendar link or a downloadable invite.
  • Execution tracking records where a draft landed and when, with a database guard preventing anything unapproved from being marked done.
  • HER OWN never sends, buys, books or shares on her behalf.

Connections and her own AI

  • Her own AI endpoint (MCP) so the assistant she already uses can read her context and use her skills.
  • Branded consent screen before any outside agent gets access.
  • Tools for reading her context, listing skills, staging an action for approval, and checking what she has approved or executed.
  • Sample prompts tailored to the skills and connections she has turned on, plus a verified state that confirms her AI has actually reached the endpoint.
  • MCP server upgraded to v0.4.0 with list_context_sources and honest metadata about what each connection is held by.

Automations and the skill runner

  • Scheduled skill runner driven by a timezone-aware automations table and a per-user run budget.
  • Tick endpoint claims automations safely, respects a kill switch, and runs up to 50 skills per invocation with concurrency limits.
  • Manual run path with enrollment checks, skill tier checks, and a per-user rate limit of 5 runs every 10 minutes.
  • Runner constructs and validates every action payload internally; nothing is sent without her explicit approval.
  • Malformed AI output is handled action-by-action; overruns cap at 5 actions and mark the run as partial.
  • Raw output capped at 8,000 characters for debugging without storage bloat.

Timezone-aware infrastructure

  • Timezone captured in her context during onboarding and used for automation scheduling.
  • Skill runs reschedule correctly using the automation's timezone, not UTC, when connections are missing.
  • Fallback to user context timezone for manual runs and degraded scheduling.

Safety and trust hardening

  • User roles stored in a separate table; admin checks happen server-side only.
  • Runner endpoints return 404 for missing or wrong TICK_SECRET to avoid discoverability.
  • Rate limiter fails closed on errors and uses a running status row to prevent concurrent bypass.
  • Model allowlist enforced on the gateway; unknown model IDs fall back to Gemini 2.5 Flash Lite.
  • ASSISTED skills excluded from the scheduled tick; they remain user-initiated only.