AI
It proposes.
You decide.
Tell it what changed in plain language and it works out what that means for the rest of your week. Then it waits.
Pending changes
Nothing changes until you accept it.
Proposed changes don't go to your calendar. They land in a panel, one line per change, and you accept or reject each one on its own. If a move would collide with something already on the board, the fix for the collision shows up as its own line too — surfaced, not quietly applied on your behalf. What you decline is remembered, so it doesn't come back next time.
Four modes
One chatbot can't do four different jobs well.
A tactical edit and a Sunday planning session are different questions. The mode you open decides how much of your week the assistant sees, and what it is allowed to change.
Quick
“Move Tuesday’s strategy block to Wednesday morning.” “Cancel the workout, my back hurts.” Single changes to the week you’re already in.
Review
The same edits, but with the whole week in view. Ask whether the week looks balanced, or what should move now that Friday got shorter.
Deep
Builds a week from scratch as a plan you read and confirm — the narrative, what each mode should get, where the hours go — rather than as live edits.
Help
Questions about Axis itself. Plumbed against a knowledge base, no tools, and it cannot change anything.
Beyond the chat
Most of it isn't a conversation.
The chat is one surface. The rest of it is small, specific, and attached to whatever you were already looking at — a button on the thing, not a prompt to compose.
Brain dump to tasks
Paste your notes into the scratchpad. They come back as discrete tasks, sorted into the same 16 modes the rest of Axis runs on, with anything that duplicates a task you already have flagged rather than quietly added.
Triage the pile
Ranks your open tasks and shows the reasoning: three deadlines stacked on one Thursday, an estimate that overruns the block it sits in, a task drifting from the goal it is filed under. Stale ones come back as a proposal to convert them into a habit or a goal, not a prompt to delete them.
Fill in the gaps
A task with no block, no due date, no goal or no mode is one the week cannot place. Assign works an attribute at a time and proposes the missing one. Status is the exception — whether something is still worth doing is a judgement, and it never guesses at it.
Duplicate detection, goal cleanup, day planning and dictation sit in the same places and follow the same rule: the proposal waits until you take it.
Your own AI
You don't have to use ours.
Building a door to somebody else's model is a strange thing to ship. You may already pay for one you trust more than whatever we would have picked — so there are two surfaces for it, and neither routes through us.
A prompt library, already loaded
Sessions for goal setting, task triage, week review, planning, and building a habit, ritual or workout — each assembled as a prompt with your real goals, blocks and history already in it. Run it wherever you like. Paste the result back and Axis parses it into records.
A folder on your machine
Data Bridge writes your goals, tasks, blocks and profile out as readable files, and watches an inbox folder for anything written back. Point a local agent at the directory — Claude Code, or anything that can read a folder — and it works on your Axis data directly. Imports preview and resolve conflicts before anything lands. It needs the File System Access API, so Chrome and Edge for now.
Where it doesn't
The week itself is built by rules.
Generating a week from your template is not a model call. It is an engine — scaling, constraints, placement, validation — so the same inputs produce the same week, and every block it places traces back to the rule that put it there. Matching tasks to blocks works the same way, down to a list of what it could not place and why. In the planning conversation every tap is deterministic; typing a constraint in your own words is the one part that needs a model.
A model that writes your whole week is impressive once and unpredictable afterwards. We would rather spend it on the parts that need judgement — reading the week back to you, arguing about what to move — and keep it out of the arithmetic.
The constitution
Thirteen rules it operates under, on every turn.
Not values on a page — a posture compiled into every request the model sees. Three of them you will notice within a week:
- It advises, it never enforces. There is no blocking, no disabling, no kill switch on you. Going off-plan is treated as information, not as failure.
- It mentions something once. If you don't act on an observation, it drops it. A suggestion you dismissed does not come back.
- It is allowed to say nothing. Not every moment needs commentary. Quiet is a valid output.
The rest are variations on the same stance: why Axis is built this way.
Credits & cost
Open about how AI is metered.
Every request is logged with what it cost, broken out by mode, and visible in your account. If you're on pace to run past your monthly budget you get a warning before you get there — and if you do run out, AI requests pause until your billing day. Everything that isn't AI keeps working.
Current budgets and per-tier details live on the pricing page.
Tell it what changed.
Axis Core is free to start. The pricing page has what each tier includes.