Your life is scattered across a dozen apps that never talk to each other. Sphere gives each part of it its own agent: health, finance, career, relationships and eight more, all writing into one memory that lives on your phone. So it can tell you the thing a silo never could: your sleep debt is why spending crept up this week. It is kind about it, it works without an account or a key, and it is iOS only.
Where this sits today, stated rather than implied. Sphere is not part of the governance stack and it is not a client for it. Nothing here talks to TokenFuse, to Wardryx or to the console, and nothing in the stack knows this app exists. The one thing they share is Engram, the memory engine, running on a phone instead of a fleet. It is a personal side project by the same people, open and tested on its own terms, and it carries none of the measurements the stack does.
This is a simulation of the app's intended behavior, compressed: one plausible week in sixteen seconds. Sphere agents file small observations into the shared on-device memory as the days pass. Midweek, the correlation engine notices two spheres describing the same week, and says so gently, with the evidence, as a pattern rather than proof.
These are real captures from the running iOS build, in an iPhone and Watch frame. Use the steps to move between spheres, tap the phone to advance, and switch the Watch between glance and quick log. The same screens a source build shows.
Each sphere is its own store with its own agent and tools. Everything they observe lands in Engram: episodic memory in one SQLite file on your phone, recalled with a query rather than an API call, so it works offline. Memories you revisit grow stronger; ones you ignore fade and are pruned. The correlation engine reads day-keyed metrics across all twelve, which is how sleep gets to explain spending.
Health, Finance, Career, Learning, Relationships, Rest, Hobbies, Travel, Mindfulness, Creativity, Home, Goals. Each sphere has its own agent, its own tools (log_water, add_expense, log_mood…) and its own corner of the shared memory. You talk to one app; twelve quiet specialists take notes.
Episodic memory via Engram, the same engine the governance stack runs, here entirely on-device. It remembers what you said last month, strengthens what you revisit, and lets the rest fade until it is pruned. Forgetting keeps recall honest and keeps agents from citing facts that stopped being true.
The reason Sphere exists. Day-keyed metrics from every sphere feed one correlation engine, so the app notices what a silo cannot: sleep explains spending, habits explain goals. It shows the pattern next to its evidence and is careful to call it a pattern, not proof.
No account, no server, no analytics pipeline. Your data lives in on-device storage, and the free AI paths never leave the phone. Deeper reflection with a hosted model is strictly opt-in: you bring your own key, and it stays in your Keychain.
The core works with zero accounts and zero setup. Quick capture (“water 2, mood 4, spent 12 on lunch”) is rule-based and runs with no model at all, like every deterministic feature; on recent iPhones, Apple's on-device models cover the AI tier for free. A key is an option, never the price of entry.
Native SwiftUI, Swift 6 with strict concurrency. 588 tests, CI-gated, no singletons in the core package. MIT-licensed and held to the same standard as the governance stack next door: clone it, read it, run it.
Sphere is the personal wing of this site: a passion project by the same builders, with the same habits. It runs the identical Engram memory engine the governance stack uses, scaled from a fleet down to a pocket, and it is built with the same habits as the rest of what we make: tested, and in the open. No enterprise anything. Just the app we wanted to carry around.
Build it from the repository, where the steps live beside the code they build.
Build from source: xcodegen generate, open it in Xcode, run it on your iPhone. No account, no waitlist. Your data never has to apply for a visa: it stays on the phone unless you export it yourself.