Optional feature disclosure
Polyglot's Japanese text loop works without an AI provider. If an operator enables the optional adapter, it uses the OpenAI API only after a learner makes one of the explicit choices described below. Declining either choice sends nothing to OpenAI.
Choosing “Plan with optional private context” sends the fixed non-diagnostic explanation item, reviewed unit forms and meanings, allowed unit IDs, and content-safety constraints. It does not send an unanswered learner response. The response is a candidate that Polyglot validates and releases only through the in-order session gate after relevant diagnostics have settled.
Choosing “Ask for advisory review” sends the preserved written response, its reviewed expected and alternate answers, assessed unit IDs, and a bounded evaluation contract. It is never used as the canonical learning evaluation and cannot directly change mastery.
Polyglot stores only validated private candidates or advisory proposals plus minimal provenance; it does not store raw provider prompts or raw transport responses. OpenAI's API data handling, retention, and account controls govern data after it leaves LoomOS. OpenAI states that API data is not used to train its models by default unless the account opts in, and that API inputs and outputs may be retained for up to 30 days except where a different eligible retention control applies. Review OpenAI's current policy before opting in.
OpenAI API and enterprise privacy information · OpenAI privacy policy
Optional speech prototype disclosure
The Sound Lab's reviewed listening recording works without a microphone or transcription provider. Microphone capture appears only when an operator enables the separate server-side speech adapter and a learner explicitly grants the consent described below. Declining leaves no audio or transcript observation.
Before the browser asks for microphone access, Polyglot records an append-only consent choice for the no-retention policy. A granted choice permits a single capture of up to 30 seconds and 5 MB to go directly to the configured OpenAI transcription endpoint. A later revocation blocks future capture; it does not erase or silently rewrite observations already saved by the learner.
Polyglot discards raw audio after the transcription request and has no audio-storage bucket, media URL, or recording replay history. A transcription suggestion remains only in browser memory until the learner explicitly saves private learner-confirmed text. The provider suggestion is not retained as a provider record; later learner corrections are preserved additively for that learner only.
A transcript is not semantic correctness, accent evaluation, or a pronunciation score. This prototype stores explicit non-retained transcription confidence and unscored semantic/pronunciation statuses; it cannot change a mastery estimate or future planner. Speech-to-text can mishear accents, dialects, short phrases, and background noise.
Once a capture leaves LoomOS, OpenAI's API data handling, retention, and account controls apply. Operators must review the current policy and configure the deployment account accordingly. OpenAI API and enterprise privacy information
Optional Vox exposure disclosure
Vox already sends an ordinary conversation turn to its configured OpenAI model when its AI brain is enabled. If you separately enable Polyglot's Vox cues, one familiar reviewed Japanese phrase and its English fallback may be added to that turn's server-built system prompt. This creates no second provider request.
This section describes only the Slice 7B cue adapter. The separate manual conversation-signal feature has its own consent and data boundary below.
Vox receives only the reviewed cue and fallback. It does not receive a proficiency level, estimate, evidence row, response, correction, transcript, profile ID, or unit ID. Polyglot receives only your authenticated identity, an opaque turn key, and the local date. It receives no Vox message, history, assistant response, topic, or tool result.
A cue-enabled response is held until the server checks exact-once use and revalidates current permission, profile state, claim, and expiry. If that check fails, the Japanese phrase is replaced with the reviewed English fallback before the response is saved or sent. Polyglot privately preserves the cue's selection provenance and delivery state, while Vox receives only the minimized cue envelope and explanation. The conversation stays in Vox. A render or skip is never learning evidence or a proficiency signal.
Optional manual review disclosure
Conversation review is a separate permission from conversational cues and is off by default. Enabling it sends nothing and schedules nothing. Each analysis requires another explicit click in LoomOS Settings → Vox.
The Loom server selects only messages you wrote to Vox during the previous 24 hours, newest first, with fixed caps of 20 messages and 12,000 characters. It sends that bounded text to the configured OpenAI model in one structured request. Vox answers, tool results, thread titles, older history, Relay/SMS, Sous, and imported Spool conversations are not included.
Polyglot stores at most three validated private proposals, the learner's accept/edit/dismiss decision, model and prompt-template provenance, counts and timestamps, and keyed opaque references to the exact supporting messages. It does not store conversation text, quotes, thread/message IDs, the raw provider prompt, or the raw provider response. Those opaque references are not sent to the browser.
Accepting or editing keeps only a reviewed private note; dismissing records only that decision. None of these actions creates learning evidence, changes mastery, adds a correction, changes a profile or learning unit, or writes to another Loom pillar. Disabling consent blocks future runs without erasing already reviewed records.
OpenAI's API/account retention and data-use controls apply after the bounded source leaves LoomOS. Review the deployment account's current controls before opting in. No nightly schedule is part of this feature.
LoomOS is a suite of personal applications that help you understand your own life — your health, habits, finances, relationships, and rhythms. Polyglot is one of those applications. This policy explains what information we collect, how we use and protect it, and the choices and controls you have. It applies to Polyglot and the LoomOS platform it runs on.
Our guiding principle is simple: your data is yours. We collect it to show you insights about yourself, we never sell it, and we never use it for advertising.
If you connect your WHOOP account, you authorize Polyglot to access your WHOOP data through WHOOP’s official API using OAuth 2.0. We want to be explicit about how that data is handled:
We do not use your personal health data for advertising, and we do not sell it. Your data is processed on a per-user basis to serve you; we do not combine it with other users’ data to profile you.
We do not sell your personal information. Information leaves our systems only in these ways:
Data is encrypted in transit. Your records are isolated per user with row-level security so that only your authenticated account can access them. Third-party integration credentials (such as WHOOP tokens) are kept server-side and are never sent to the browser. No system is perfectly secure, but we take reasonable measures appropriate to the sensitivity of health data.
We retain your data for as long as your account is active or as needed to provide the service. You are in control:
To make any of these requests, contact us at privacy@loom.guru.
Depending on where you live, you may have rights to access, correct, export, or delete your personal data, and to object to or restrict certain processing. Because LoomOS is built around showing you your own data, much of this is available to you directly in the app; for anything else, contact us and we will help. We will not discriminate against you for exercising these rights.
Polyglot is not directed to children under 16, and we do not knowingly collect personal information from them. If you believe a child has provided us information, please contact us and we will delete it.
We may process and store your information in the United States or other countries. By using Polyglot, you understand that your information may be transferred to and processed in countries with different data-protection laws than your own.
We may update this policy from time to time. When we make material changes, we will update the effective date above and, where appropriate, notify you. Your continued use of Polyglotafter an update means you accept the revised policy.
Questions, requests, or concerns about your privacy? Reach us at privacy@loom.guru.
← Back to Polyglot