How it works
We're still writing
how it thinks.
ORCA already runs. Explaining it properly — every signal, every decision, every point where it stops and asks — takes longer than building it did. That write-up is in progress. Here's what it will cover.
Reading the room
The 8+ signals ORCA weighs before it does anything — time, location, calendar, weather, budget, inventory, habits — and how they get resolved into a single situation.
The memory graph
How preferences, routines, services and past decisions compound into a personal graph, and why ORCA never starts a task from scratch twice.
Deciding what it's allowed to do
Risk scoring, trust thresholds, and the exact line between a task ORCA completes on its own and one it stops to ask you about.
Execution, in the open
What actually happens when a task runs across connected services — what's logged, what's reversible, and what you see while it happens.
Learning from being wrong
How an accepted, edited, or rejected action changes the next one — the feedback loop that makes the graph sharper over time.
Want a chapter sooner, or one we haven't listed?
Tell us what to write first