Local first
If computation can happen on your device, it should.
Lower latency, greater resilience and less data leaving your control. The cloud remains useful. It just shouldn’t be the automatic answer.
Powerful software doesn’t need to live in a datacenter.
Default Local builds intelligent software designed to run close to the people and data it serves.
Local by default. Efficient by design.
Cloud when it actually helps.
Remote infrastructure
APIinferencedatabase
Your deviceNetwork optionalNetwork required
Your data
Local computeRequest / response
Persistent state
A change in direction
A different
default.
For years, software has moved in one direction.
More servers. More round trips.
More dependencies. More compute.
We think some of the most useful software of the next decade will move the other way.
Onto your devices. Closer to your data. Available offline. Private by architecture.
Efficient enough to remain quietly running in the background.
We call that Default Local.
Architecture
Same work.
Less in the way.
The network is a useful tool.
It doesn’t have to be part of every thought.
- 01device
- 02network
- 03API
- 04inference
- 05database
- 06network
- 07device
Principles
Different defaults
produce different systems.
Do less work
The cheapest computation is the computation you don’t perform.
Process incrementally. Reuse existing state. Recompute only what changed.
Small before large
Bigger isn’t automatically better.
Classification, retrieval and extraction often need specialised systems, not giant models. Use expensive reasoning when reasoning is necessary.
Architecture before brute force
A better representation can eliminate enormous amounts of computation.
Efficiency begins with system design rather than hardware.
User ownership
Models change. Infrastructure changes. Your information should endure.
Keep user data understandable, portable and independent from any one runtime.
Context Engine.
Software that remembers
the world around you.
We’re building a local cognitive system that continuously maintains a structured model of your world.
Instead of repeatedly giving an AI everything it might need to know, Context Engine is designed to maintain that understanding incrementally: what changed, when, where information came from and how confident the system should be.
Explore Context Enginesourcesobservationslocal state ●understanding
Why now
Intelligence is moving onto the device.
CPUGPUNPU
The hardware is here.
The architecture is catching up.
Modern phones and personal computers contain remarkably capable CPUs, GPUs and neural accelerators. At the same time, useful models are becoming smaller and more specialised.
That changes the architecture of intelligent software.
Tasks that once required sending data to remote infrastructure can increasingly happen millimetres away from the user.
We think that shift is only beginning.
Before adding more
The Default Local test.
Can it run locally?
Can a smaller system solve this?
What changed?
Can this be rebuilt?
Does it actually need to leave the device?
Simple questions produce surprisingly different software.
Compute closer.
We build software around the computer you already own — and use remote infrastructure when it earns its place.
About Default Local