AI you can identify, question, and overrule.

This notice explains FlowSight's intended purpose, AI boundaries, data flow, and human-oversight controls under the EU AI Act. It is written for the people using the product and the organisations deploying it.

Current position

FlowSight is intended as human-supervised work-environment decision support. It is not intended to evaluate a person's job performance or make employment decisions. Using it for those purposes may bring a deployment within the AI Act's high-risk employment category.

Last reviewed: 10 September 2026

What the system does

The product separates local observation, dashboard calculations, and optional generative assistance. Those layers do not have the same data or purpose.

On-device classification

The desktop agent can classify work context locally. Raw screenshots and detailed screen context are not designed to become a manager-facing cloud feed.

Deterministic analytics

The dashboard calculates measures such as meeting load, focus-block coverage, and context switching from available workspace signals.

Generative assistance

The AI Coach and eligible report summaries use a third-party model hosted through Microsoft Azure. The model explains supplied data; it does not make decisions for you.

Product safeguards

These controls implement the transparency and human-control principles that are relevant to FlowSight's current intended purpose.

  • People are told when they are interacting with AI and when a summary is AI-assisted.
  • Coach answers cite the FlowSight metric or user-supplied document used for a factual claim.
  • AI output is advisory, may be incomplete or wrong, and must be reviewed by a person before action.
  • The Coach is restricted from ranking people or recommending hiring, dismissal, promotion, discipline, compensation, or task-allocation decisions.
  • FlowSight does not provide workplace emotion recognition, biometric categorisation, or social scoring.
  • Uploaded documents and past chats are treated as untrusted context, not as instructions that can override product safeguards.

Uses FlowSight does not support

Do not use FlowSight output as the basis for hiring, dismissal, promotion, discipline, compensation, work allocation based on personal behaviour, or any other decision that materially affects a person's employment. Do not use it for covert monitoring, emotion recognition, biometric categorisation, or inference of sensitive characteristics.

Flow, focus, meeting, and workload-pressure indicators describe incomplete work patterns. They are not measures of effort, ability, health, intent, or individual performance.

Deployer responsibilities

  1. 1.Tell workers and, where applicable, their representatives what is collected, why it is used, and who can access it.
  2. 2.Establish a lawful basis, complete any required data-protection or fundamental-rights assessment, and respect local employment and works-council rules.
  3. 3.Keep a qualified person responsible for reviewing outputs, challenging errors, and stopping use when the context is inappropriate.
  4. 4.Do not repurpose FlowSight output for individual performance scoring or employment decisions.
  5. 5.Give people a practical route to question data, correct context, and report a harmful or misleading output.

Accuracy, records, and reporting concerns

AI output is probabilistic. Citations improve traceability but do not guarantee a correct conclusion. Users can delete Coach conversations, challenge the underlying context, and choose not to act on a recommendation. We maintain proportionate security, change-management, risk-review, and incident-response processes for the AI features we provide.

Report a misleading, discriminatory, unsafe, or unexpected output to privacy@flowsight.site. Include the date, feature, and enough context for us to investigate without sending unnecessary personal data.

Legal context

The EU AI Act is risk-based. Its transparency rules apply from August 2026. The Commission's current implementation timeline places Annex III employment-system requirements from December 2027. Classification always depends on the actual intended purpose and deployment context, so customers must assess their own use with qualified counsel. This page is product information, not a conformity declaration or legal advice.