Platform & API data: no sale, no sharing, no retention
SynapsAI Cloud does not sell, rent, license, monetize, or share your API or platform processing data — including prompts, parameters, model inputs, inference outputs, embeddings, audio, images, or video — with advertisers, data brokers, analytics vendors, model-training partners, or any third party for their own commercial purposes. We do not keep the content of processed inference payloads after your request completes. The only exceptions are strictly temporary handling required to run the job:
We do not use your API or platform data to train models, build behavioral profiles, or feed marketing analytics. Infrastructure subprocessors (including Nebius, Cloudflare, and Better Stack — see our Subprocessors list) receive only what is technically necessary to execute your workload and are contractually bound not to access, copy, or retain your payload content.
See Section 3 for technical detail and Section 7 for how this differs from optional marketing-website analytics.
1. Client Data Ownership & Controller Status
Clients retain absolute ownership, title, and intellectual property rights over all input data payloads, prompt strings, parameters, files, and outputs processed by the service. SynapsAI Cloud acts strictly as a data processor executing compute tasks at the precise, automated instruction of the client.2. Permitted Data & Licensing Requirements
Clients must possess all necessary legal rights, permissions, and licenses to process any data they input into SynapsAI Cloud infrastructure. The service must not be used to process data in violation of regional data privacy laws, intellectual property rights, or international export controls.3. Transient Processing Architecture & Zero-Retention Commitments
SynapsAI Cloud uses a transient processing architecture for inference payloads and API processing. Under this model, prompt content, model inputs, and generated outputs are processed to fulfill each request and are not retained as persistent customer data afterward. This zero-retention commitment applies to inference payloads and API processing, not to separate platform records such as model weights, deployment metadata, operational logs, or billing metrics — and not to optional marketing-website analytics collected after cookie consent. In short: we process your data to fulfill the API request, then discard it. We do not archive prompts, outputs, or media for later use, and we do not disclose them to third parties except infrastructure providers acting solely to run your job under strict contractual controls.- Standard Inference Payloads: Text prompt arrays and standard tensor structures submitted for inference are processed entirely in volatile memory (RAM/VRAM) and are never intentionally written to persistent disk media. They are not logged, exported, or retained after the request ends.
- Audio & Video Preprocessing: Temporary files generated exclusively for audio/video normalization or preprocessing are written to isolated, ephemeral, single-tenant filesystem partitions. These temporary assets are programmatically wiped and purged immediately upon the completion or failure of the compute pipeline job.
- Generated Video Assets: Outputs resulting from video generation pipelines are maintained temporarily within volatile memory on our host infrastructure for a maximum lifetime of 10 minutes from creation. This window allows the client to securely fetch the artifact via standard egress API protocols. Clients may trigger an immediate, programmatic purge of these assets prior to the 10-minute expiration by hitting the explicit delete endpoint.
4. Permanent Model Architecture Caching
To achieve low cold-start latency and sustain high token throughput, SynapsAI Cloud permanently caches model weights and related compilation artifacts on specialized, high-throughput storage volumes.- These model files are locked strictly to the client’s tenant account and specific deployment configuration.
- These files are retained permanently to facilitate instant initialization but can be fully, irrevocably destroyed by the client at any time by executing a delete request for that specific model deployment through the management control plane.
5. System Metrics, Operational Logs & Diagnostic Data
Standard operational system logs, metric counters, reverse-proxy access logs, monitoring alerts, and execution traces are aggregated to ensure platform health, network optimization, security review, and granular billing calculations. By architectural design, these records are strictly isolated from client execution environments and do not include raw input request payloads, prompt contents, string fragments, or generated model response tokens. The following operational metadata may be retained for limited periods as needed to operate the platform:
If a rare diagnostic event requires temporary capture of non-payload technical context (for example, a redacted stack trace or deployment identifier), that data is used solely to investigate and resolve the incident and is not used for model training, advertising, or resale.
6. Security Incident Notification
In the event of a verified physical or structural infrastructure security breach affecting customer data segments, SynapsAI Cloud will implement its incident containment procedures and notify affected clients within the timelines mandated by applicable statutory laws.7. Marketing Website Analytics (Separate from Platform Data)
The synapsai.cloud marketing website may use third-party analytics tools (Google Analytics 4 and Microsoft Clarity) only after you consent via our cookie banner. This website analytics is separate from SynapsAI Cloud inference processing:
We do not use marketing-site session recordings, GA4 data, or Clarity replays to train models, access customer inference payloads, or profile authenticated platform users. No API inference data is ever sent to Google Analytics, Microsoft Clarity, or other marketing analytics tools.