Memory Constellation
Every fact bimi knows about you, drawn as stars.
Identity, preferences, commitments, corrections, episodic moments — categorized, dated, and connected. Tap a star to read or revise. Reject one and bimi forgets.
Appalogue · indie Mac-first software studio
The AI assistant
with a memory you can read.
Start free — no API key, no account.
It learns your work and remembers everything — on-device, editable, yours to audit. Bring your own Claude or GPT if you want.
Already invited? Sign in to the tester portal →
Built by one person in Florida Zero telemetry SDKs Public roadmap
The idea
An assistant that knows you.
And knows your work.
Every AI assistant forgets you the moment the tab closes. That's the part we refuse to accept. bimi is built on the opposite idea: a memory you can audit that compounds for years, and an assistant that adapts to your work — your machines, your clients, your cases — with the receipts to prove everything it did.
The platforms are building ambient assistance into the OS. Good. We're building the half no platform ships: memory that's actually yours — on your device, in plain sight, editable, and accountable.
Appalogue exists to build that path.
-
Memory you can see
Every fact bimi remembers is in a constellation you can browse, edit, and prune. No black box.
-
On-device first
When your Mac can run a capable local model, a local brain — a llama.cpp- or MLX-served open model sized to your Mac — handles memory, extraction, and classification on-device. Chat runs on the model you pick. When that's a hosted model, the facts relevant to your question are sent to it — see Privacy for exactly what's shared and what's redacted.
-
Adapts to your job
Network engineer, attorney, founder — bimi meets you there. Drop a switch config and it types out the VLANs, logs the device, and answers against it weeks later. Every profession gets its own skills, tools, and memory shapes.
-
Proactive, not pushy
Signals become memories silently. Reflection happens in the background. bimi speaks when it matters.
Flagship product
bimi — your native agent.
Conversational AI, with memory. Memory-rich. Privacy-respecting. Built natively for macOS today — iPad, iPhone, Watch, and home shipping in waves.
bimi — Building Intelligent Meaningful Interactions.
Voice that knows who's talking
Diarization + speaker ID, on-device.
bimi tells you apart from your spouse, your kids, your coworkers — using locally-stored voiceprints. Raw audio never leaves the device. (Coming in the relevance-engine wave.)
Signals → Memory → Reflection
A three-stage loop that runs while you sleep.
Calendar events, location pings, music, app activity become signals. Salient ones promote to memory. Reflection synthesizes patterns. The end result: bimi gets to know you without ever needing to interrupt.
Hub-and-spoke across your devices
macOS is the hub. The rest are spokes.
Heavy lifting on your Mac (or the M-series chip you already paid for). Lightweight spokes on iPhone, iPad, Vision Pro, Watch, and TV. One memory, one identity, every screen.
Smart hybrid routing
Local for the easy stuff, cloud only when it earns it.
When a capable local model is installed, background work — classification, extraction, memory — runs free on your Mac's on-device model. Conversations run on whichever brain you pick: fully on-device for privacy, or your own Claude/GPT key. With a hosted key, the memories relevant to a turn are sent to that provider (credentials and financial data redacted; names and personal facts included). Every routing decision is named.
Native, end to end
SwiftUI, FoundationModels, CryptoKit, MetricKit.
No Electron. No web wrapper. No telemetry SDKs. Crash diagnostics use MetricKit. Sandboxed App Store core; notarized Power Helper for the SSH / shell power users.
Learns your machines as it works
Connect once — bimi remembers the box, the login, and the key.
The first time bimi reaches a host over SSH, it records the machine and the login that worked, and imports your key into the encrypted Vault. Next time you just say “connect and check it” — bimi resolves the machine, the user, and the credential itself, with nothing to re-enter. Every box you touch fills in your fleet automatically, and the secret never leaves the Keychain — the registry only ever holds a handle. (New in beta.)
Join the bimi beta
macOS 26 Tahoe · Apple Silicon · Apple Intelligence required · iPhone/iPad/Watch/TV/Vision Pro spokes shipping in waves
Where this is going
The longer arc.
None of this is in the build you can try today. It's here so you can see the shape of the thing and tell us if we're heading the right direction.
Cluster work, for real
- Sub-agent fleets — spin up N parallel workers per goal, merge into one memory thread, ask questions across the whole thing.
- Long-running operators — give bimi a goal once, let it run for hours or days, ask for clarification when stuck.
- Learned procedural skills — teach it your Monday review workflow once, watch it run autonomously next week.
- Job Kits — tell bimi your profession once; it loads the tools, skills, and memory shapes to match.
Voice, the way you'd use it
- Speaker recognition — knows who said what in a room, addresses people by name.
- Voice character with warmth — not a monotone TTS; restraint, humor, and the patience of an assistant you'd actually want to talk to.
- Household awareness — different responses for you, family, a guest. Privacy stays on-device.
Wearables & sensors
- Smart glasses — first-person video + audio grounding what bimi sees with what you see.
- Pendant capture — always-with-you audio, speaker-attributed, on-device transcription.
- Sensor fusion — calendar + photos + voice + glasses + pendant collapsed into one context.
Proactive without being annoying
- Presence layer — context awareness that knows when to speak and when to fade.
- Adaptive rituals — weekly check-ins that actually learn you, dismissal feedback loops.
- Predictive timing — the right tool at the right moment, derived from calendar + location + pattern.
Memory you can see
- Memory constellation — dark-sky UI where memories are stars and links are constellation lines.
- Reflection engine — bimi thinks between turns; distills lessons during idle.
- Multi-vector retrieval — embedding + temporal decay + category boost + entity linking, all at once.
Trust, in public
- Memory Inspector — see exactly what bimi remembers, edit, delete, audit every change.
- Trust posture dashboard — privacy nutrition labels, on-device ratio, model + prompt change impact in plain English.
- Conversational permissions — mid-conversation asks for sensitive actions; declines remembered per source.
A privacy manifesto
The defaults are the product.
Anyone can claim "privacy-first" on a marketing page. We try to mean it in the code.
-
01
On-device whenever your Mac can run a capable model.
When a capable local model is installed, bimi runs it on your Mac for classification, extraction, and memory work. Cloud providers are opt-in, listed by name, and every routing decision is visible. When you pick a hosted chat model, the memories relevant to your question are sent to it — credentials and financial data redacted, names and personal facts included so the answer is accurate.
-
02
Your voice is yours.
Voice is push-to-talk: the mic opens when you press it, and speech recognition runs locally. Raw audio is never persisted or transmitted. Voiceprints — if you opt in, once speaker ID ships — will be stored as sensitive memories, encrypted with your device key.
-
03
Memory is auditable.
Every fact bimi remembers has provenance: where it came from, when, and how confident we are. You can inspect it, edit it, hold it in review, or delete it. Forever.
-
04
No telemetry SDKs. Ever.
We don't ship Mixpanel, Amplitude, Sentry, Crashlytics, or any third-party analytics. Crash diagnostics are MetricKit, attached to your bug reports — only when you submit one — and end-to-end encrypted to us.
-
05
Beta reports are encrypted to us, not Anthropic.
When you send feedback, the payload is encrypted with X25519 + ChaCha20-Poly1305 on your Mac and decrypted only on our server. We don't read it through any third party.
-
06
Open about what we can't do.
Where we use the cloud — long-form reasoning, embeddings on big corpora — we say so. Where we don't yet meet our own bar, the in-app trust badge shows yellow and we tell you why.
The shop
Built by one person.
That's the point.
Appalogue is a one-person Mac-first software studio. Every line of Swift, every privacy decision, every visual token is made by someone who has to live with the consequences.
No shareholder pressure to add tracking. No growth team padding "engagement" with dark patterns. No quarterly directive to leverage AI in a way that grates against the rest of the product. Just opinionated software I'm proud to run on my own machine.
bimi is the flagship — and likely the only product Appalogue ships for a while. That's deliberate. Memory you can actually trust is a big enough problem to fill a career.
— Steven Imbimbo
founder · Appalogue LLC · Florida, USA
Join the bimi beta
First impressions deserve first builds.
We're letting in a small group of testers this season. macOS 26+ on Apple Silicon. You'll get the latest signed build, a private feedback channel, and a real conversation about what you'd want bimi to do next.
Application received
We've got it.
We read every application personally — usually within 48 hours. You'll get a one-time setup link at the email you gave us once you're approved.
Already invited? Head to the beta access page →
Shipping cadence
Recently shipped to the beta.
Pulled live from the build manifest. We ship to the beta multiple times a week.
Loading the latest builds…