AI policy

Asten today

Asten’s checks do not run any AI model, on your computer or in the cloud. When Asten marks that Mira’s hair is black in chapter 14 after chapter 2 settled it as auburn, that mark comes from a deterministic rule comparing facts you confirmed. The same project gives the same result every time.

Release 0.0.1b adds an optional assistant to the desktop app. The assistant panel answers from Asten’s own tools by default. If you turn on AI providers, choose one, and consent, it can also call that provider’s model to explain a mark, search your canon, summarise a source, or propose a fact for you to confirm. It has ten tools that read or prepare and no tool that edits your manuscript; a request to write, rewrite or continue prose is refused before any provider is contacted, and a reply that reads like new narrative is withheld. What the assistant sends lists exactly what leaves your computer, when, and to whom.

Asten does not generate prose, and it never will. The tagline is a design rule: let AI be your editor, not your writer. Your words are black ink; Asten only ever writes in red.

Principles every AI feature meets

The assistant in release 0.0.1b is the first optional assisted feature. Further ones, such as proposing facts from your prose for you to confirm, are plans, not promises, and none has a date; see the roadmap. Every AI feature follows these principles:

• Editor, not writer. AI in Asten never writes, rewrites, or continues your prose.
• Proposals only. AI output arrives as a proposal: a suggested fact, a possible contradiction, a question. Nothing takes effect until you accept it.
• Never silent canon. An AI proposal never becomes a confirmed fact on its own. Only you can confirm it; the assistant has no tool that does.
• Local first. Deterministic rules come first, then models that run on your computer (through Ollama or any OpenAI-compatible address on your machine). Nothing leaves your computer unless you decide it should.
• Your provider, your key, your consent. Asten sends text to a cloud AI provider only if you choose that provider, use your own account and key, and consent to that provider. Consent is tied to the provider’s address, and the interface shows which provider a request goes to.
• A complete switch-off. You can turn every AI feature off in one place (Settings, AI providers, Allow AI providers). Off means no model runs and nothing is sent. Offline mode allows only a model on your own machine, and a spending cap of zero means no call.
• No training on manuscripts, ever. Your manuscript will never be used to train or improve a model, by Verne or by any provider Verne chooses.
• You can see what is sent. Before a request leaves, you can open it exactly as the provider will receive it, with your key replaced by a placeholder. A local ledger records each call: when, which provider and model, how many tokens, and an estimated cost.

How Verne builds Asten

Verne may use AI tools in its own work, such as writing code or drafting text. That is outside this policy, with one firm exception: your manuscripts are never part of it. Verne never receives your manuscript, so it cannot be fed into any tool Verne uses.

How this policy changes

If this policy changes, Asten will tell you in the interface first, and this page will show the new date. Consent you gave is tied to a version of the notice: a new version asks again. Turning on any AI feature is always your choice.

For the thinking behind this policy, see Asten and AI and AI as editor, not writer. Questions: info@vernesystems.com.