Brains & Models
Orb separates who it is (the personality you create) from what powers it (the model). You can swap the model freely; the personality is constant. This is why Orb feels like one assistant no matter what’s running underneath.
The tiers
On-device (always available)
A compact model runs directly on your phone. No internet, no account, no key. It’s the floor that’s always there — fast for everyday things, and the reason Orb works on a plane or with no signal. It’s deliberately honest about what it can and can’t do locally, and will offer to hand bigger jobs to a stronger brain when one is connected.
Free APIs (chained)
Add one or more free cloud providers and Orb rotates through them automatically: when one hits its limit, it falls through to the next. Stack a few and you have a lot of free capability. Setup: Free APIs & Chaining.
Your own server
Run the Orb server on a PC for the most capable tier — including doing real things on that computer. Setup: Self-Hosting.
How Orb stays consistent
Whichever tier answers, Orb is fed the same personality you set up. The model is background infrastructure. So:
- Switching brains doesn’t change who Orb is — only how much it can do.
- A weaker brain doesn’t get a different attitude; it just has fewer abilities, and says so.
- You can let Orb pick the best available brain automatically, or pin one in Settings.
Falling through, gracefully
The chain is ordered by capability and availability. A turn flows down the chain until something answers:
your own server → free API #1 → free API #2 → … → on-device floor
If a server or API is unreachable or rate-limited, Orb skips it for a short cooldown and uses the next one — you just get an answer. The on-device floor never gets skipped, so a request never dead-ends.
The point: you set this up once and then forget it exists. Orb just works, and it appears to fix itself because the fallbacks are automatic.