Your direction through the atlas
The Atlas is the world. The Passport is your record of it. The Compass is your direction through it — a 32-dimension taste model learned from every drive.
The Compass Instrument
A specimen taste model across 32 dimensions, learned over 40 drives. Its strongest axes are Scenic, Mountain, Viewpoint, Remote, Nature, Quiet.
Drag to turn · scroll to zoom · right-drag to panHold an axis to name it
- drives learned from
- 0of 40
- strongest signal
- —
- model settled
- 0%across 32 axes
A specimen, not a person: the model above starts at nothing and is fed forty completed drives. Every axis is one of the 32 dimensions the product carries. Cool where it is still unsure, warm where it has committed — the needle rests on whatever is strongest so far, and stops hunting as the vector settles.
A working compass, derived from real drives
The instrument above is not a drawing — it is your taste model made tangible. The needle points to your strongest signal. The arcs show learned preference strength across all 32 dimensions. Every figure derives from the committed registry, not a questionnaire.
32 dimensions
Read from the same constant the taste tools use — no hand-typed list to drift.
Learned, not declared
The model moves with every drive — route events update the vector automatically.
Account-bound, not surface-bound
One taste profile powers the app, Byway AI, MCP, and (soon) phone and car.
The 32 axes of taste
Every dimension is a place category the engine scores. The list below is generated from the single constant the favorites panel and taste tools read — a retyped list would be a second copy free to drift.
- scenic
- viewpoint
- coast
- beach
- mountain
- nature
- waterfall
- historic
- cultural
- food
- local gems
- hidden gems
- hidden towns
- family
- wildlife
- adventure
- art
- photography
- sunset
- sunrise
- quiet
- popular
- free
- paid
- easy
- strenuous
- pet friendly
- accessible
- remote
- urban
- educational
- romantic
Categories match FAVORITE_CATEGORY_CHOICES in src/lib/favorites/osmLink.ts. Order is registry order.
Nothing learned from others outranks what's learned from you
This is a bound, not an absence — it constrains every future capability. Pooling the shape of taste (which categories co-occur) is a fact about taste, not a guess about a person, and it is the only honest answer to sparsity. But it ships as intent, not present behavior. The bound survives its own roadmap.
No cross-user ranking
Your taste vector is never blended with another traveler's. The model is stored against your account, not the surface.
Shape pooling is intent
Learning which categories co-occur is a fact about taste, not a person. It is the only honest answer to sparsity — and it ships under 'Building next'.
Explicit choice leads
Your onboarding picks always rank first. The learned model only fills gaps — it never overwrites what you declared.
One taste profile, everywhere you drive
The Compass is not a surface — it is the model that powers every surface. The architecture stores taste against the account, so a surface is a client, not a copy.
App plan
Byway weights every byway toward what you love — and learns the rest as you go.
Byway AI
Conversational drive requests resolve through your taste vector.
MCP
Assistants call the same taste profile via the MCP tool surface.
Native app shares the same account-bound taste vector.
Tesla / CarPlay / AA surfaces read the same vector.
Pick what you love. The model learns the rest.
Onboarding takes 60 seconds. Your explicit choices seed the vector; every drive refines it.