1. Role as an Infrastructure & Execution Layer
Platform vs. Provider: Sinas.co provides the managed control plane, serverless runtime, state persistence, and orchestration engine. We do not develop or train underlying foundation Large Language Models (LLMs).
Model Neutrality: Customers connect their own choice of foundation models, fine-tuned weights, or third-party provider API keys (e.g., OpenAI, Anthropic, Mistral, open-source models). Sinas routes, executes, and persists state based strictly on Customer configuration.
2. Data Privacy & Zero Model Training Commitment
No Training on Customer Data: Sinas does not use Customer prompt inputs, execution payloads, database states, system instructions, or output logs to train, fine-tune, or validate any public or commercial foundation models.
Transient Payload Processing: Data passing through our serverless execution environments and agent orchestrators is processed transiently to deliver services. Persistent storage (such as vector state or key-value caches) is strictly isolated and encrypted at rest using AES-256.
EU Sovereignty: Sinas defaults to primary hosting infrastructure located within the European Union (Germany / Ireland), fully compliant with GDPR requirements.
3. Transparency & EU AI Act Compliance Support
Sinas enables enterprise clients to comply with international regulations, including the EU Artificial Intelligence Act (EU AI Act), by embedding structural compliance measures directly into our managed platform:
Auditability & Traceability: Sinas records deterministic telemetry, execution traces, tool call history, and latency metrics for agent workflows to ensure system behaviors can be fully audited.
System Logging: Detailed execution logs allow developers and compliance teams to inspect agent reasoning chains, API inputs, and automated output states.
Content & Output Identification: We support standard metadata headers and structured JSON outputs that enable our customers to attach machine-readable watermarks or transparency disclosures to AI-generated user-facing outputs.
4. Safety Controls, Guardrails & Human-in-the-Loop
While primary responsibility for application-level content safety rests with the Customer, Sinas provides built-in platform guardrails:
Human-in-the-Loop (HITL) Triggers: Sinas orchestrators support native suspension points where automated agent workflows pause for human review and authorization before executing high-risk tool calls or external API actions.
Resource Hard Limits & Circuit Breakers: To prevent runaway loops, resource exhaustion, or unintended recursion, Sinas enforces strict timeout limits, memory caps, and configurable maximum call depths on all serverless execution environments.
Abuse Monitoring: Sinas monitors platform-level network traffic and resource consumption to prevent malicious uses of our infrastructure, including network scanning, automated exploitation, or deployment of malicious code.
5. Security & Isolation Architecture
Tenant Sandboxing: All custom Python or Node.js execution environments execute inside microVM/containerized sandboxes with restricted network interfaces.
Secrets Management: Customer API keys and external access tokens are encrypted in transit and at rest using hardware-backed key vaults, accessible only by the isolated runtime instance at execution time.
6. Continuous Evaluation & Inquiries
As global AI regulations and technical standards evolve, Sinas regularly reviews and updates its platform architecture, security controls, and ethical guidelines.
For inquiries regarding Sinas’s AI safety practices, audit logging, or compliance documentation, contact us at support@sinas.co.