AI Use & Data

What the writer sees

When configured, the writer receives a step screenshot, guide title, step number, clicked label, current step title, page origin/path and up to two neighbouring step labels on each side, along with instructions for writing. It returns a title, body and reviewer notes. The guide is saved first; empty steps can then be sent automatically, without another per-guide AI consent screen.

The intended hosted route is OpenRouter to Google Gemini. The code supports other configured models, compatible endpoints and fallback writers. The exact deployed model, underlying serving provider and fallback configuration are unknown, needs Yash. An AI failure may send the same input to a configured fallback. It does not send your account password as a prompt field, but any secret left visible in a screenshot can still be sent.

Training and retention are separate questions

Asterly does not use customer content to train its own models. That does not establish what a provider retains or how its service is configured.

OpenRouter states that prompt/response storage and its use of those inputs are opt-in, while request metadata is retained. Underlying providers have separate training and retention policies and routing controls. We have not verified Asterly’s account switches, contracts or restricted routing.

Google’s direct Gemini API terms distinguish paid and unpaid handling: paid inputs/outputs are not used to improve its products; unpaid content may be used for improvement and reviewed by people. Google’s abuse-monitoring page states a 55-day retention period for prompts, context and outputs. The applicability of those direct-API terms to the actual OpenRouter serving route needs confirmation. We do not promise that all requests are training-free or immediately deleted.

Provider sources checked on 25 October 2026: OpenRouter data collection, provider routing and logging, Gemini API terms, and Gemini abuse monitoring. These describe provider policies, not proof of the settings or contracts on Asterly’s account.

Redaction and choosing whether to use AI

Sensitive data in recordings: use a test account and made-up data. Keep passwords, payment details, health records, private messages and other people’s information off-screen. Screenshots can capture anything visible. Review every screenshot and step before sending or publishing.

The inspected code contains automatic pixel masking of some emails, phone numbers, password fields, marked private areas and likely avatars, plus an editor control for masking a selected area. This is a partial implementation, not verified coverage or a guarantee about the installed extension. Visual material in frames, canvas, images or changing pages can escape detection. Masking in the hosted editor may happen after AI has already received the first screenshot.

A dependable pre-upload manual redaction workflow and per-guide AI opt-out are not verified as available. There is no separate AI-off choice in the send flow we inspected. Do not send confidential content expecting an opt-out later. If AI is unsuitable, contact hello@asterly.co before uploading; no private-processing exception is promised.

Review and future features

Generated copy is a draft. Check it against the workflow, remove sensitive information and correct errors before publishing. No AI system is infallible. Customer content must not be used to make high-impact decisions about people through this service.

Speech, narration and music are outside the reviewed data flow. No voice or music provider is authorised by this page. Before introducing one, we need to identify the provider, inputs, retention, rights and consent requirements and update this notice. For access or deletion requests, use Data deletion & contact.