Local-first, not local-only
Use the browser and the user's device when they are capable. Connect to remote services when the task genuinely needs them.
Useful intelligence
should be closer.
Lwati is designed around a simple idea: not every piece of intelligence, computation or source code should have to travel to a remote server before a person can get useful work done.
Deterministic before generative
Search, calculation, SQL, tests and scientific runtimes should solve deterministic problems directly. AI is reserved for reasoning and creation.
Portable work
Source code, research artifacts and assessments should remain exportable and inspectable.
Graceful capability
A weak GPU should not make the editor, SQL engine or research workspace useless. Different device classes should get different capability paths.
DESIGNED FOR REAL CONDITIONS
Powerful software should not assume permanent broadband or unlimited inference budgets.
Local runtimes, cached assets and ordinary deployment targets matter anywhere — and they matter even more where connectivity, hardware and cloud budgets are constrained.
Local by default.
Connected when useful.
Portable by design.
Connected when useful.
Portable by design.
BUILD WITH AI / RUN WITHOUT IT
Intelligence can be capitalised into software.
When AI helps create deterministic workflows, those workflows can run later as normal code. More users do not automatically need to mean more model calls.