Agents and Saverick LM
AI agents that build with you,
in your language.
Not a chatbot bolted on the side. Saverick agents use the platform itself: they place blocks, run the simulator, read sensor data and prepare a deployment, then hand control back.
What it does
- Tutor agent explains concepts and errors, and asks questions back
- Build agent turns a spoken or typed idea into blocks you can inspect and edit
- Debug agent runs the program in Sim, reads telemetry and suggests a fix
- Teacher agent drafts lesson plans and summarises who is stuck and why
- Hardware agent checks wiring and pin maps, then sends the parts list to the Store
- Guardian agent helps moderators keep spaces for young learners clean
At a glance
- Status
- In development
- Target
- H1 2027 · LM Edge alpha
- Editions
- Desktop and web, with limits
- Languages
- All supported packs
In detail
Sim first, always
Agents test every program on the digital twin before proposing it for real hardware.
A human approves
Flashing firmware, moving a robot, buying parts or posting publicly needs a person's confirmation.
Teachers in control
Teachers set how much help agents give, see every action in a log, and can switch agents off.
Runs offline
Saverick LM Edge is a small quantised model that runs on learner laptops with no connection.
What it won't do
Stated here so nobody discovers it after buying a kit.
- Saverick LM is adapted from open-weight base models — we are not training a frontier model.
- Edge runs a smaller model than Pro, so complex planning needs a school server or the cloud.
- Local-language quality depends on data we are still collecting, with consent, from native speakers.
Coming soon
Get told when it ships
Saverick AI is coming soon — Saverick is currently in pre-launch.