AI is probabilistic
Datanito AI products may generate incorrect, incomplete, inconsistent, or outdated information. Output should be treated as generated assistance, not guaranteed fact. Important work should be verified against reliable sources and appropriate experts.
Models may vary
Products may use different first-party or third-party models depending on task, availability, region, quality, latency, cost, or configuration. Model behavior can change over time. Where a specific model materially matters to the user experience, we aim to expose that information in the product or documentation.
Agents can take actions
Agentic features may call tools, APIs, MCP servers, search systems, databases, code environments, or other services. We design for scoped permissions, visible tool activity, approvals for consequential actions where appropriate, and records that help users understand what occurred.
Memory and knowledge
Products may use conversation history, workspace context, retrieved documents, or user-configured memory to provide continuity. Users should be able to understand the relevant scope of retained context and control or remove it where the product supports that capability.
Evaluation and monitoring
We aim to evaluate important AI behaviors using test cases, regression checks, tool-call validation, traces, latency and cost monitoring, and failure analysis. No evaluation can prove a system is error-free; evaluation is an ongoing engineering process.
Human review
Datanito products are not a substitute for required professional judgment. Users are responsible for human review when decisions are high-impact, regulated, irreversible, or could materially affect rights, finances, health, safety, employment, security, or legal obligations.
Feedback
We use product feedback to identify failures, improve interfaces, and strengthen evaluations and safeguards. Report material issues to [email protected].