Trust Center

AI Service Terms

Explains AI-specific responsibilities, model-provider routing, input/output handling, safety limits, and evaluation use.

Effective: 2026-06-16

AI Service Terms

Effective date: 2026-06-16

These AI Service Terms supplement the Terms of Service for features involving AI models, model routing, agent workflows, tool execution, retrieval, document processing, evaluation, and generated outputs.

1. Inputs and outputs

Users provide prompts, files, messages, API requests, tools, instructions, and context as inputs. The service may return model outputs, tool results, summaries, code, structured data, images, embeddings, or other generated materials. Users are responsible for inputs and for reviewing, validating, and deciding how to use outputs.

2. Model routing and providers

The service may route requests to first-party infrastructure, third-party model providers, customer-selected providers, bring-your-own-key providers, or fallback providers. Routing may depend on model availability, latency, cost, safety, product configuration, plan limits, resilience, and user settings. Provider names and subprocessors may be disclosed in the Praxis Trust Center or product configuration.

3. No automatic training assumption

User original content is not automatically treated as training data. Model training, fine-tuning, evaluation, or human review uses must be based on product settings, customer instructions, provider configuration, enterprise agreement, or explicit disclosure. Operational logs, diagnostic metadata, routing metrics, aggregate analytics, and safety signals may be used to operate, secure, debug, and improve the service.

4. Output limitations

AI outputs can be inaccurate, incomplete, outdated, biased, unsafe, or unsuitable. Outputs may resemble third-party material. Users must not rely on outputs as legal, medical, financial, safety, or other professional advice. Users must apply human review where outputs may affect rights, safety, compliance, or material decisions.

5. Safety and prohibited use

Users must comply with the Acceptable Use Policy and must not use AI features to evade safety systems, create harmful content, impersonate others, perform unauthorized security activity, generate regulated professional advice without proper oversight, or make unlawful automated decisions.

6. Customer controls

Customers are responsible for configuring model choices, API keys, retention settings, access controls, workspace permissions, and downstream use. Customers using bring-your-own-key or external providers are responsible for ensuring their provider terms, data processing terms, and transfer mechanisms are appropriate for their use case.

7. Monitoring and enforcement

We may use automated and manual systems to detect abuse, security risks, policy violations, and operational failures. We may limit, block, or reroute requests where needed to protect users, providers, the platform, or the public.

AI 服务条款 | 知行云 · 知行云 Praxis AI Cloud