A local-first AI learning companion for precision education. Fikira helps learners think, practice, and build confidence without turning learning into answer-copying.
Free beta. Mac 26.10.4 is signed, notarized, and Gatekeeper verified. Windows 26.10.4 is signed by Frame the Change LAB through Azure Trusted Signing. Android 0.1.0 is a scaffold test APK for course-code learner companion validation, not a Play Store release yet. Learning work stays on your device by default. Fikira is part of a ProSocial AI research agenda focused on access, teacher support, and meaningful study help.
Fikira Mac 26.10.4 and Windows 26.10.4 are the current releases. New in 26.10.4: multiple-choice practice checks no longer put the right answer in the same position every time, and Fikira tells you when this computer can run a larger Qwen3 model. 26.10.4 also fixes the problem that stopped 26.10.3 from opening; if you installed 26.10.3, download 26.10.4. Also included from 26.10.2: teachers can generate course codes again, and course changes and the encrypted workspace backup save correctly when signed in. Also included from 26.10.1: emailed sign-in and reset codes now work reliably, a teacher's workspace opens only for that teacher and is stored encrypted, and only signed-in teachers and training providers can create courses. Security settings for the app's local connections were tightened. The first screen is School or college or Workplace training. Teachers, students, and parents stay in school. Providers open a catalog-first desk. Participants enter a private organization code.
The Mac app is signed with Apple Developer ID and accepted by Gatekeeper as a Notarized Developer ID app outside the Mac App Store.
The Windows installer is signed by Frame the Change LAB through Azure Trusted Signing. Version 26.10.4 is the current signed Windows beta release.
Learning conversations and saved work stay on the computer by default. Optional account, sync, feedback, and update services use the internet only when needed.
Fikira is moving toward smoother shared-device profiles, profile restore, and cross-device child access while keeping local use simple.
The first Android scaffold is available as a test APK. It is for validating course-code join and learner study actions before Play Store signing and distribution.
Fikira is not trying to be a chatbot with school branding. It is being shaped around learning science, responsible AI, and practical classroom and family use.
Fikira is designed to handle the technical setup for you. No command lines. No separate app installs. A short profile setup helps personalize the experience.
Click the download button for your system. Mac users: choose Apple Silicon (M1/M2/M3) or Intel. Once the DMG downloads, open it and drag Fikira to your Applications folder.
On Mac: version 26.10.4 is signed with Apple Developer ID, notarized by Apple, and Gatekeeper verified. It is distributed outside the Mac App Store. If macOS blocks the first opening, use the in-app first-open steps below.
On Windows: download and run the installer. The 26.10.4 Windows installer is signed by Frame the Change LAB through Azure Trusted Signing.
Ready to use in about 60 seconds.When you open Fikira for the first time, it checks whether the local AI model is ready on your device. If the model needs to be downloaded, Fikira handles this completely on its own.
You will see a progress bar while Fikira downloads the Qwen model that fits your computer. This happens once, and the download size depends on your device. After setup, the core course tools can run privately on the device. Updates and optional online services still need internet access.
On your first visit, choose School or college, or Workplace training. School or college then offers Teacher, Student, or Parent. Workplace training then offers I run programs, or I have an organization code.
Return visits go to the last world and seat. If you are in the wrong room, use the small link to switch.
Setup takes under two minutes.Fikira runs the AI entirely on your device, so it needs enough storage and memory to run the local model comfortably.