The shipping log
No testimonials yet: MetaRun is pre-launch. What we can show instead is the pulse. Every entry below is a real ship date from the commit history of a product the studio uses on its own six live apps every day.
- ListingMILOMCPMonetizationSite
Import your fastlane metadata folder, and MILO knows its next move
The Listing workspace now imports fastlane deliver's metadata folder in one previewed change: every locale diffed against the live listing, locked or invalid fields skipped with named warnings, and nothing written until you confirm. Export produces the same folder back as a zip, and both run as MILO/MCP tools too, so an IDE agent can migrate a listing straight from your repo. Tool results also now carry suggested next actions: after creating a product MILO is pointed straight at pricing and localization, a drifted change points back to a fresh preview, and a locked listing points at the exact editability check. The Settings connect-a-client command for Claude Code now defaults to `--scope user`, so the MCP server follows you into every project instead of only the folder you ran it in. Subscription listings now project as a monetization bay in the mission feed (plan, product id, cadence, status, and current US price) instead of a wall of text, scoped to exactly what you asked for and never squatting between a reply and its ENGAGE slate. Bulk subscription and IAP repricing no longer fails in Monaco, Bangladesh, or Samoa (Apple carries no price points there; they're now skipped with a named warning instead of erroring), and reverting a still-scheduled price change now correctly cancels it instead of failing. Adding a listing language no longer sets you up for a rejected submission: Apple requires a privacy policy URL per language and never inherits it, so creating a locale now carries your existing URL over as a visible, warned diff row (verify the page serves that language, or ask MILO to set a language-specific link), the Listing workspace exposes the privacy URL fields per language, and the submit-for-review preview now blocks up front, naming exactly which languages still miss one.
- MILO
Faster MILO, visible thinking
MILO's replies now stream in live, with a small cognition window showing its rolling thoughts and an ops telemetry line for every tool call while it runs. Missions also start faster: the deck pre-loads your fleet so MILO skips the ritual first lookup.
- MILO
MILO finishes multi-step missions
When you ENGAGE a proposed change, MILO now picks the mission back up on its own and continues to the next step instead of waiting to be asked. Proposal slates also wait politely for the reply to finish streaming before they materialize.
- MILOMonetization
Create products conversationally
Subscription groups, subscriptions, and in-app purchases can now be created through MILO end to end, with duplicate product-id sweeps before anything is staged and pricing chained right after. Same-tool proposals gained a batch ENGAGE ALL, and export compliance declarations became one guided step.
- MILOSite
The deck goes mobile
The Command Deck and the voice link now work properly on phones, so you can run your store by talking to MILO from anywhere. Support tickets can be filed in-app, by asking MILO, or from your IDE over MCP.
- BillingSecurity
Plans, the trial, and the Power Plant
Metered plans landed with a 14-day no-card trial, a billing deck with a live usage reactor, and honest fail-closed metering on every AI surface. The same week brought the operator onboarding ceremony, the metarun.dev domain, a full pre-launch security hardening pass, and the Terms and Privacy pages.
- DecksASO
Three new command decks
Reviews became the Comms Array (triage-sorted, analytics that double as filters), Releases became Launch Control (versions, TestFlight triage, submit-for-review), and ASO got its own Command Deck with an hourly rank sentinel, a promo-text scheduler, and per-territory star ratings.
- StudioPricingMILO
MILO learns to design, and to price
The Screenshot Studio became fully MILO-operable with server-side rendering, so screenshots can be designed, rendered, and published to App Store Connect without the studio tab even open. The pricing engine landed too: PPP, Big Mac, and Spotify index strategies snapped to Apple's tier catalog, shared by the deck, chat, voice, and MCP.
- StudioMILO
The Screenshot Studio and the voice link
The studio was rebuilt end to end: 3D device mockups, AI-generated captions, per-locale translation, and one renderer shared by preview and export. The chat page became the Command Deck, including a real-time voice link: talk to MILO, watch proposals materialize, press ENGAGE.
- MILOSafety
Memory and the universal undo
MILO gained persistent memory (visible and editable in Settings) and MetaRun gained a universal change history: every applied change from every surface, dashboard, chat, voice, or IDE, is recorded with a before/after diff and almost all of them revert in one tap.
- Decks
The manual decks
The Listing and Pricing workspaces landed as full manual decks: locale matrix editing, categories, age ratings, accessibility declarations, territory-grid repricing with equalization overlays. Everything MILO can do stays fully doable by hand, behind the same preview-and-confirm contract.
- MILOPlatform
The registry triples
The Apple mutation layer grew from 12 to 31 preview/apply operations and MILO's tool registry from 34 to 91 tools in one push: pricing points, reviews, TestFlight feedback, featuring nominations, in-app events, and more. One registry now feeds the dashboard, the chat agent, and the MCP server alike.
- MILO
MILO moves in
MILO went from a launcher widget to the home of the product: a full-page chat with persisted conversations and AI-generated titles. The same week set the launch scope honestly: Apple-only first, Google Play deferred until it can be done properly.
- Origin
First commit
MetaRun begins as Hysa Tech's internal tool for running its own six-app App Store fleet, because doing it by hand in App Store Connect was the bottleneck. The plan from day one: every capability built tool-callable, so an AI operator could eventually run all of it.