Where this is going.

The workshops

Rooms that read a text you paste and show you its shape, with no account, no key, and nothing uploaded. All three are open. The writer's room finds who recurs and where, the rhythm of the sentences, and draws the cast as strands that meet and part. The philosopher's room reads where an argument turns and which of your terms changed the company they keep, quotes the claims either side of a turn exactly rather than writing them for you, and can hand what it found to your canvas. The documents room reads who is quoted, who is only ever talked about, when things happened, and what sums are named.

Open is not finished, and the honest gap is in the documents room: it reads a document and cannot yet keep one. The finding it exists to produce — these two files you filed separately disagree with each other — needs more than one text in scope at a time, which is the workspace item further down this list. Until that lands it is a reading rather than a case file.

A first-run experience designed on purpose

The way the app is arranged today reflects the order things were built rather than how they relate — the same idea appears in two places, and one screen holds two different ways to do the same thing. This is a pass over the whole information architecture, including what someone sees in their first sixty seconds.

Nobody has ever experienced a new account except by building and testing one themselves. Every interface bug found in early August was invisible from the inside.

Carry a reading across to the canvas

What a workshop finds should be able to become part of your graph — the terms, the people, the links between them — with a review step where you see exactly what is proposed and can strike anything you do not believe before it lands. The philosopher's room does this today: it hands you its claims as a paste you read before you accept it. The other two rooms cannot.

It shipped as one room's own path rather than as plumbing all three can use, which is both the honest description and the reason this is still here. Lifting it out is the work, and it is what stops each new room having to build its own way across.

Pricing

A free tier that is genuinely useful — your nodes, lenses and entries are never capped, and a full export is free at every tier. Above that, you pay for the work that actually costs something to run: a model reading or rewriting across your graph. Billed through a merchant of record who handles tax properly wherever you are.

The seller is set up and subscriptions work; the numbers are not decided, and publishing invented ones is the single hardest thing to walk back.

Usable without a mouse, and without seeing the graph

Every surface that isn't the canvas — docs, library, cards, review, settings, the panels — reachable and operable by keyboard, announced properly by a screen reader, and legible at the contrast levels a person actually needs. For the graph itself, a parallel way through the same structure: an outline of lenses, nodes and their relations that can be read and navigated linearly, because a force-directed diagram cannot be made meaningful by describing where circles are.

Partly done and never verified. Lens and scale controls are real radiogroups with accessible names and arrow-key navigation, and the Mandala's refraction has a keyboard path — but nothing has been tested with an actual screen reader, and the custom theme editor lets anyone pick colours the app cannot guarantee any contrast for. Naming it here rather than claiming it: an accessibility promise made before an audit is a promise made without evidence.

A workspace that holds more than one document

Everything that reads a text today reads the one in the box. Working across a pile — six papers, forty filings, a folder of articles — means the app has to know that a document is a thing with a name you can come back to, rather than whatever is currently pasted.

It is the one item on this list that is architecture rather than work, and the room that most needs it is the one not yet started. Doing it early would be building a foundation for a building nobody has drawn.

Choose the model per task

One setting currently governs every operation — writing entries, weaving connections, making cards. Different jobs want different models, and the same mechanism is what lets you bring your own key.

Bring your own key

Use your own provider account from your own browser. Deliberately browser-only: keys are not held server-side.

Custody of other people's API keys is not something to take on before the legal entity exists to be accountable for it.

Edit back through the same channel you ask through

Asking your graph a question and following each claim back to the node it came from is live. The other direction is not: restating a claim across forty nodes at once, or rewriting a whole lens in a different register, without opening forty editors.

It is the first thing that could write model-authored text into your graph at scale, and the review step that marks such text as a draft has to cover it before it ships rather than after.

Mobile

Reading, reviewing drafts, and running cards travel well. Authoring is a desk activity and this will not pretend otherwise at launch.

Curation and sharing

Publishing a lens for others to fork, with the provenance intact — including a visible warning before an action that would re-parent somebody else's work.

Capping how much you can make

Nodes, lenses and entries are never limited by a plan, and a full export is free at every tier. Text in a database costs almost nothing to keep, so a cap on it would not be pricing — it would be leverage applied to your own thinking, by the tool you keep it in. What genuinely costs money is a model doing work across your graph, and that is the only thing metered.

Scoring your sources for you

No credibility rating, no bias meter, no verdict on a publication. What the tool will do is show you the shape of what you have gathered — how many separate places a claim came from, how much of it is attributed rather than asserted, where two of your own documents disagree — and stop there. The difference matters: the first tells you what to think, and the second tells you what you are holding.

Resolving disagreements between lenses

When two readings of the same idea conflict, the tool holds both and shows that they differ. It will not merge them, rank them, or pick a winner. A disagreement between two careful readings is information, and flattening it is the specific thing this is built to avoid.

Holding your API keys on our servers

Bring-your-own-key will stay in your browser. Server-side key custody is a liability that wants a real legal entity and an audit behind it, not a convenient afternoon.

Full authoring on a phone

Not because it is impossible, but because promising it before anyone has tried to build a lens on a phone would be a promise made from a guess.

Training on your work

What you write is yours. It is not a corpus.