Frequently asked. Specifically answered.

Fill out the form on the home page. We read every application, usually within 48 hours. Approved testers get a one-time setup link by email — no password is ever sent over email.

Open-ended. We're treating the beta as the path to the App Store release, not a 30-day window. As long as we're shipping breaking changes, the beta stays open.

No — the beta is free. After beta, bimi is a flat subscription: $14.99/month or $129.99/year (save 28%), with a one-time $299.00 lifetime option — and a free tier stays available. We won't surprise existing testers with a bill; early supporters get looked after.

Two reasons. First, the rough edges need to be discoverable individually — we want to be able to read every report and reply. Second, some features (voice, multi-agent fleets) require us to onboard testers in waves so we can isolate signal from new bugs.

It's usually a "not yet," not a "no." We re-evaluate the queue every two weeks as new build capacity opens up. You can re-apply, or email hello@appalogue.com if your situation changes.

You click the magic link, set a password, and land on a download page that hands you a signed Apple Silicon .pkg. The first launch shows the Gatekeeper warning — bimi is still ad-hoc signed; Developer ID notarisation is queued behind App Store enrolment. The install page walks you through right-click → Open.

It depends on the chat model you choose. When a capable local model is installed, background work — classification, extraction, memory — runs on your Mac, with local embeddings for retrieval. Answers run on the chat model you pick: a hosted provider (Claude, GPT) receives the memories relevant to your question, with credentials and financial data redacted but names and personal facts included so the answer is accurate. An on-device chat model keeps everything on your Mac. Every cloud call is named in the sidebar.

No. Deleting a memory leaves a tombstone — the same utterance won't silently re-create it. If you want it back, restate it.

Both get marked "in review" in the Memory Inspector — we don't pick a winner silently. An in-conversation surfacing pass that asks you when it's natural is on the roadmap. See the decay section in how-memory-works for the longer version.

Nowhere off-device. Speech recognition runs locally, and voice is push-to-talk in the current beta. The wav lives in RAM during the utterance, becomes text + signal, and is dropped.

For Meeting Mode, the m4a is written to a sandboxed cache so bimi can transcribe a multi-hour session. Today the default retention applies to that file (a user-facing retention picker is landing in a near-term wave). The transcript + summary become memories; the audio is purged per policy. No audio is ever transmitted to a server.

If you enable it: yes, to your CloudKit container, with the encryption posture the platform provides for CloudKit. Appalogue does not operate a sync server. iCloud sync is opt-in.

None. No Mixpanel, no Amplitude, no Sentry, no Crashlytics, no Bugsnag, no PostHog, no Firebase. Crash diagnostics use MetricKit and are attached only to bug reports you submit. See privacy.

• macOS 26 Tahoe or later
• Apple Silicon (M1 or later) — Intel Macs are not supported (Apple Intelligence is Apple-Silicon-only)
• Apple Intelligence enabled in System Settings (Settings → Apple Intelligence & Siri)
• 16 GB RAM recommended; 8 GB works with cloud fallback
• ~250 MB disk for the app; memory storage scales with use (typically < 100 MB)

Intel Macs aren't supported. The memory architecture leans on Apple's on-device Foundation Models framework (which is what Apple Intelligence is built on), and Apple Intelligence is Apple-Silicon-only by design. We hate stranding users, but the experience without the on-device model is meaningfully worse for everyone, so there is no Intel path going forward. If you're on Intel and want a heads-up when something changes, write us at hello@appalogue.com.

Shipping in waves. macOS is the hub. The spokes share the same memory and identity. See hub-and-spoke for the topology.

No. bimi's recommended brain is a free hosted model — no key, no account, no setup (with a fair-use daily limit). Bring your own Claude/GPT key only if you want a specific model, or switch to fully on-device for privacy.

Your conversations run on whichever brain you pick: bimi's free hosted model out of the box, fully on-device (Apple Foundation Models) for privacy, or your own Anthropic / OpenAI / OpenAI-compatible key. Background work — classification, extraction, embeddings, memory — runs free on your Mac's on-device models.

Yes. In Settings → Providers & Models, pick “Your own” (or open Advanced) and add your key — it lives in the macOS Keychain. Routing decisions are shown in the sidebar so you can see when the key is being used.

A separately notarised download for capabilities that don't fit inside the Mac App Store sandbox — SSH, shell, and similar. The core bimi.app ships sandboxed via the App Store; the Power Helper is opt-in for power users.

No, not today. The source might open in parts over time — particularly around the memory architecture — but the application binary is closed-source while we get the privacy posture right and the build stable.

Short version: bimi keeps memory and voice on your Mac, has structured auditable memory, is a native Mac app instead of a tab, and shows you every cloud call by name. ChatGPT is a better general-purpose LLM product; bimi is a different kind of thing. Long version on /compare.

Ollama runs models. bimi is an agent that uses models (including Ollama, if you point it there) inside a memory + reflection loop. If you want to wire your own agent on top of Ollama, do — we'd just rather build the agent and ship it.

No. We're presenting at Cisco Live (the conference), and that's a common confusion. Steven is an independent solo developer. More on /about.

Not today. The shape of the product depends on the structure being small. If that changes, we'll say so on the about page.

hello@appalogue.com — for everything. Use "urgent" or "security" in the subject line if you need prioritisation.

At /press. Logos, screenshots, founder bio, and copy-paste company description for journalists.