Albena

How it works

Personal AI you run yourself.

Albena runs on your own hardware with local models. She is shaped around you, your household or your business, connects to the systems you already use, and never makes a change without your approval.

1

Local first

Everyday requests are handled on hardware you own, using local models and Apple on-device intelligence. That keeps them fast and private, and they keep working when the internet does not.

  • Choose a fast or deep lane per task
  • Cloud model only when a task needs it and you allow it
  • You are told what leaves your home
Architecture overviewYour voice, screens and devices connect to Albena running on hardware you own. Inside, a lane router chooses local models, which use memory you control. Every request passes an approval gate before reaching your connected systems such as smart home, work tools and cloud services. An optional cloud model is used only if the task needs it and you allow it.Hardware you ownVoiceSay “Hey Albena”Screens & keyboardTouch and typing tooYour devicesSpeakers and displaysLane routerFast or deep, per taskLocal modelsPlus Apple on-device intelligenceMemoryYours to view, edit and deleteApproval gateReads pass. Changes wait for you.Smart homeLights, locks, mediaWork & commsMail, chat, ticketsCloud & ITInfrastructure, identityCloud model (optional)Only if needed and you allow itArchitecture overviewYou talk to Albena. On hardware you own, a lane router picks local models and uses your memory, then an approval gate decides what may reach your systems. An optional cloud model is used only if allowed.Hardware you ownYouVoice, screens and your devicesLane routerFast or deep, per taskLocal modelsPlus Apple on-device intelligenceMemoryYours to view, edit and deleteApproval gateReads pass. Changes wait for you.Cloud modelOptional, only if allowedYour systemsHome, work, cloud and identity
Every request passes through the same path. The dashed box is optional and off unless you allow it.
2

You approve every change

Reading is instant: ask about anything connected. Anything that changes something is proposed to you first, shown exactly as it will run.

  • High-risk actions need a change record and your PIN
  • Full audit log of what was asked and done
  • Rollback information saved with each change
  • Per-user permission tiers for the household or team
Approval flowA request is planned. If it only reads, it is answered instantly. If it changes something, it is proposed to you, you approve or decline, then it runs once and is recorded in the audit log with rollback information. High-risk changes also require a change record and your PIN.You askVoice or textAlbena plansPicks the right laneReadAnswered instantlyChangeProposed to youYou approveOr declineAction runsOnce, as shownHigh riskAdds a change recordand your PINAudit logWhat, when, and how to roll backApproval flowA request is planned. Reads are answered instantly. Changes are proposed to you, approved or declined, run once, and are logged with rollback information. High-risk changes also need a change record and PIN.You askVoice or textAlbena plansPicks the right laneReadAnswered instantlyChangeProposed to youYou approveOr declineAction runsOnce, as shownHigh riskChange recordand your PINAudit logWhat, when, how to roll back
Reads are instant. Changes always go through you.
3

Memory you control

Albena keeps her own memory of your people, routines and preferences so you do not repeat yourself. It lives on your hardware first, and you can see and delete any of it.

  • Learns from usage overnight and proposes improvements
  • You approve; she never deploys changes to herself
  • Delete one memory or all of them, any time
Memory and learningEveryday use feeds a nightly review, which creates proposals. You approve them. Memory about people, routines and preferences is kept on your hardware and you can view, edit and delete it at any time.Memory, on your hardwareEveryday useRequests and routinesNightly reviewFinds useful patternsProposalsImprovements to reviewYou approveNever self-deploysPeopleWho is whoin your lifeRoutinesWhat happensand whenPreferencesHow you likethings doneYour controlsView, edit,deleteMemory and learningEveryday use feeds a nightly review which creates proposals you approve. Memory of people, routines and preferences stays on your hardware, and you can view, edit and delete it.Memory, on your hardwareEveryday useRequests and routinesNightly reviewFinds useful patternsProposalsImprovements to reviewYou approveNever self-deploysPeopleWho is who in your lifeRoutinesWhat happens and whenPreferencesHow you like things doneYour controlsView, edit and delete
Learning is a proposal, never an automatic change.

Want the evidence behind these claims? Read the trust center and the control mapping. Ready to try it?

Join the waitlist