Meet Mira — your personal AI, built around you.
Mira is a personal AI assistant that runs on your device when it can, and on a remote runtime when it can't — so you get a fast, capable assistant without handing every conversation to the cloud by default.
Mira
Running locally on this device
A personal AI assistant, built to actually know you.
Mira is designed to be the assistant you come back to — one that keeps context across conversations, adapts to how you work, and gives you a say in how it runs.
A conversation that continues
Mira isn't a fresh chat window every time. It holds the thread of a conversation and pulls in relevant facts you've shared before, so you're not re-explaining yourself.
An assistant, not a search box
Mira reasons through problems with you — planning, writing, coding help, working through a decision — rather than returning a single flat answer.
Built to run close to you
On supported devices, Mira runs its model locally instead of routing every message through a distant server by default. When that isn't possible, a remote runtime takes over.
Built around control, not lock-in.
Most assistants make you trade privacy for convenience. Mira is built so that trade-off isn't automatic.
Local AI
On supported devices, Mira runs a model directly on your hardware — fast, and without a round trip to a server for everyday chat.
Privacy-focused architecture
Your conversations and memory are scoped to your account and isolated at the data layer — not just hidden behind UI, enforced where the data lives.
Flexible AI providers
Mira is the product; the model underneath is an implementation detail. It runs on local inference today, with a remote runtime and additional providers as options.
Cross-platform by design
Built to work the same way whether you're on Desktop, Mobile, or Web — though exactly what's available can differ by platform.
Memory & personalization
Mira remembers relevant preferences, ongoing projects, and decisions you've explicitly shared, and recalls them when they're actually relevant.
User-controlled experience
You're not locked into one model or one provider. As more options ship, choosing how Mira runs stays in your hands.
One request, the best available path.
Mira decides how to handle each request based on what's actually available — starting local, and falling back only when it needs to.
You
Your request
Mira
Decides how to route it
Local Runtime
Runs on your device when your hardware supports it.
Remote Runtime
Handles the request when local isn’t available, with usage limits on the free plan.
Cloud AI Providers
Coming soonOptional third-party providers for extra capability.
P2P AI Compute
Coming soonInference via authorized peer devices.
One assistant, wherever you work.
Mira is designed to run across Desktop, Mobile, and Web. Capabilities — like whether Local AI is available — can differ by platform and device.
Desktop
A native app for Windows, macOS, and Linux — where Local AI is best supported today.
Mobile
A mobile app is planned, with device-appropriate AI depending on what your hardware supports.
Web
A browser-based version is planned for quick access without installing anything.
Start free. Upgrade when you need more.
Pricing details are still being finalized — here's the shape of what each plan includes.
Free
Free
Everything you need to start using Mira day to day.
- Core Mira experience
- Local AI where supported
- Remote Runtime access, with usage limits
- Ad-supported
Paid
PlannedComing soon
For people who want more headroom and capability.
- Higher usage limits
- Cloud AI access, when available
- More advanced capabilities as they ship
- Ad-free experience
Privacy as an architecture choice, not a promise.
We'd rather explain how Mira actually handles your data than make sweeping claims about it. Here's what's real today.
Local inference stays local
When Local AI handles a message, the prompt goes to the model running on your device for that step — not to a third-party AI provider.
Your data is isolated to your account
Conversations and memory are scoped to your account and enforced at the data layer, not just hidden in the interface. Other users cannot query or see your data.
Secrets are never remembered
Mira is designed not to store passwords, API keys, or other credentials as ordinary memory, even if you share them in a conversation.
Authentication is verified, not assumed
Access to your account is checked server-side on every request — the app itself is never treated as the source of truth.
Where Mira is headed.
These are directions we're building toward, not features available today. We'll label them clearly as they ship.
More AI providers
Additional cloud model providers as optional choices, so Mira can tap into more capability when you want it.
A P2P AI network
Inference through authorized peer devices, as another option alongside local and remote runtimes.
More powerful local models
A broader catalog of local models and an easier way to choose which one runs on your device.
Extensions
A permissioned way to add tools and integrations to Mira, rather than a fixed feature set.
Deeper personalization
Richer long-term memory and context, so Mira's understanding of your work keeps improving.
Start building your relationship with Mira.
Mira is in early access, starting on Desktop. Sign up to get notified as new platforms and capabilities roll out.