AI Providers & Data Use

This notice supplements the Privacy Policy, Terms of Use and, for business processing, the Data Processing Agreement. It describes the current beta's actual provider route, not planned integrations. SYNPL is operated by WORKSTORIES, INC.

1. The current architecture

Your selected workspace, source index, personal Wiki and working memory are kept on your Mac. Cloud AI is different: Mac -> api.synpl.com on Hetzner in Germany -> OpenAI, with responses returned through the service. The SYNPL application server receives and relays the relevant request and response without retaining their bodies. It separately stores account, subscription, usage and device records.

TLS protects the AI network transport. This does not prevent the receiving SYNPL server or OpenAI from processing content. Do not confuse it with the end-to-end encrypted Mac/iPhone relay, whose encrypted content SYNPL cannot read. Apple receives push tokens and encrypted notification payloads for delivery. Relay metadata and operational records are still processed.

2. Current cloud provider and functions

OpenAI is the only active cloud AI provider in this release. Anthropic is not currently used. A future provider will require an updated disclosure and the relevant permissions and contractual safeguards before receiving covered information.

Local transcription of supported voice messages uses local Whisper-based processing and is distinct from real-time cloud voice. Recordings imported through Plaud are processed only when that source is connected; this does not add a general call-recording feature. Model availability and task routing can change within the disclosed provider arrangements; no particular model or answer quality is guaranteed.

3. Minimization and the limits of masking

The application limits extraction to selected sources and uses a configurable source-history window, seven days by default. Its text pipeline uses pseudonyms for names, addresses and phone numbers and detects categories of secrets before indexing. Task context is reduced before cloud processing. The local request record and activity controls help you inspect and limit what goes to the model.

These controls do not guarantee that all data is anonymous or safe to disclose. Context can identify a person; patterns can miss a secret; manually entered prompts, search terms, spoken text and audio can reveal details. Treat outgoing data as potentially personal and confidential, and connect only information you are entitled to process. Local workload/pressure indicators and crisis-marker matching stay local as described in the Privacy Policy; this does not prevent a separately authorized task from using underlying source excerpts.

4. OpenAI retention and training

SYNPL does not use Customer Content to train general-purpose models or opt it into OpenAI's model-training programs. Applicable OpenAI API arrangements distinguish no-training use from retention. For supported responses we set store=false, but this is not a zero-data-retention approval and does not eliminate all provider processing or storage.

OpenAI normally retains abuse-monitoring records, which can contain inputs and outputs, for up to 30 days. Legal or safety requirements and feature-specific processing can create exceptions. Some capabilities also involve temporary application state or caching. SYNPL has no zero-retention arrangement represented for this release and does not promise that all provider data is erased after exactly 30 days. Deleting local data does not remotely erase provider records.

The applicable reference is OpenAI's API data controls. Its consumer ChatGPT settings are not the privacy settings for SYNPL's API integration. We remain responsible for our own provider configuration and commitments; a provider policy does not authorize us to weaken an accepted processing restriction without the required steps.

5. Authorization and Google information

Cloud tasks, including scheduled digests you enable, must remain within the purpose and sources you authorize. Connection permission, approval to send a message, cookie consent and authorization to process information with cloud AI are different choices. Where affirmative data-transfer consent is required, it must be collected before that transfer, through the relevant in-product disclosure. These documents alone do not collect it.

For Google, the enabled permissions are Gmail read/send and read-only Google Calendar. Google-derived context sent through SYNPL to OpenAI remains covered by Google's Limited Use restrictions even when pseudonymized or aggregated. We use it only for the disclosed, appropriate user-facing functions and permitted exceptions, not advertising, sale, credit scoring or shared-model training. The detailed undertaking and revocation process are in Privacy Policy Section 5.

Sending communications requires your specific approval, including any approved future schedule. Selecting cloud AI does not authorize the model to send unrelated communications or change its own permissions.

6. Other service recipients

Hetzner hosts the German intermediary and server records. Amplitude receives text-free interface-event analytics through SYNPL's metrics proxy; Stripe handles payments; Apple handles push delivery; Webflow hosts the relevant website. These are not additional cloud model providers and do not automatically receive the same data as OpenAI. The Privacy Policy specifies their functions. Your independently connected accounts have their own provider arrangements.

OpenAI processing involves the United States. Other recipient locations depend on the relevant service arrangement. The German server does not guarantee EU-only processing, and masking does not replace a required international-transfer safeguard. Applicable business processing and subprocessors are governed by the DPA.

7. Controls, responsibility and changes

Use source selection, the history window, pause, local mode, the local model-request record and erase controls to manage exposure. Disabling product analytics is a separate setting in "About"; it is currently an opt-out control, not a universal consent mechanism. Local mode does not promise that login, billing, synchronization, analytics or phone services are all offline. To stop a connected source completely, pause/disconnect it and revoke its provider access as appropriate.

The Terms' beta disclaimers and liability limits apply without a separate cap here. We give no guarantee of perfect anonymization, absolute security, continuous provider availability or correct output. You manage device security and lawful use; we retain our own data-protection, processor and approved-action obligations. Neither a personal laptop nor a third-party model removes those duties.

Material changes follow the notice and acceptance rules in the Terms and Privacy Policy. New data recipients or purposes requiring consent will not be authorized merely by continued use of an unrelated feature.

Contact: WORKSTORIES, INC., 2093 Philadelphia Pike #2517, Claymont, DE 19703, United States. Email: support@synpl.com.