Products
What we build.
Life, organized on one canvas.
Free to start
One infinite canvas for tasks, goals, habits, notes and calendar. Built for a brain that wants clarity, not more tabs.
- 01Comprehend. See your whole week before you decide anything: energy, priorities, the time you actually have open.
- 02Coalesce. Calendar, tasks, notes and habits that finally talk to each other. Two dozen connectors, one surface.
- 03Agency. You drag it. You arrange it. You decide. Canvas, timeline, board, matrix, whichever fits the moment.
- 04Intuit. Kaya watches your rhythms and surfaces the right task at the right moment. Not a chatbot.
Paid tiers add an allowance of AI actions for a flat monthly fee. Not a token meter.
Hand your agent a request. It comes back solved.
From $5 · first 3 free
An MCP server that works a request out end to end. It breaks the request into the parts you actually have to answer, finds a real tool for each, and runs the route. What works gets kept, so the same request in the same spot comes back the same way next time.
- Routes, not guesses. A request comes in, your agent gets a proven path it can run.
- Deterministic. The context decides the steps, and the steps are fixed. Repeatable and auditable.
- It compounds. Every verified solve is stored and reused, so the next one is faster and cheaper.
Packs: $5 for 20 asks, $22 for 100, $125 for 1,000, or $75/mo.
A model that becomes yours.
From $999, one-time
A personal model that runs on your own machine and keeps adapting to how you work. Buy it once, own the install. No token meter.
- Runs on your hardware. In private beta now, and capable from day one.
- Adapts to you over time. The model itself changes, it's not a memory note pasted into a prompt.
- Offline when you want it. One toggle seals every outside call.
Pre-order now; your card isn't charged until it ships.
A model that keeps learning after it ships.
By request
A self-modification system that changes the model's actual weights, attaches adapters, and relearns. It can train itself on a skill or a workflow directly, down at the layer level. That's not a skill file, and it's not agent tweaking. The model rewrites itself.
- Changes real weights. Not a frozen base with an adapter bolted on. It retrains specific layers, edits facts in place, distills, prunes, and merges, and attaches adapters on top.
- Trains itself on skills and workflows. Show it the behavior and it learns the behavior, in the weights, not a prompt and not a stored skill file.
- Neurogenesis. It grows new capacity when it runs out, the way a brain adds neurons, instead of only reweighting what's already there.
Also from the lab
More from the model line: Noesis, the full reasoning model, not out yet. And Liminal, a model trained on one piece of your software, built by request.
See the models →