We Decode Communication Without Building a Surveillance Archive
Decodeme analyzes text to surface organizational risk and never builds a hidden archive of employee communication. We strip personal identifiers before storing anything, and the original message is discarded after 72 hours — your History and Ask Decodeme run on that de-identified version. No facial or voice biometrics, no AI trained on your data. Privacy is the architecture — which is also what keeps the model legal where camera- and voice-based tools are not.
Four Principles, Built In — Not Bolted On
Every principle below is an architectural decision, not a setting you have to find. Privacy by design is what lets an organization monitor communication risk without becoming the thing employees fear.
De-identified by default
Names, emails, phones and locations are stripped from the analysis before it is stored, and the original message — along with the identifiers it arrived with — is deleted 72 hours later. The organization never gets a searchable archive of employee messages, only team-level aggregates.
No biometrics, ever
Decodeme infers signals from written text only. No facial expression, no voice tone, no physiological data — the categories that carry the heaviest legal and ethical risk.
Your data never trains our models
AI models are never trained on customer communication. Your messages are inputs to a signal, never training data for someone else's product.
Aggregated, consented, controlled
Organizational reporting is anonymized to the team level. Individuals consent, control their own data, and access is role-based.
Why Text-Only and Ephemeral Is a Legal Advantage
The EU AI Act prohibits emotion recognition in the workplace — but Article 5 limits that prohibition to biometric data. Text-based analysis falls outside it. Camera- and voice-based emotion tools do not.
Anchored in Recognized Standards
- International standard for psychosocial risk at work
- Decodeme reports are designed to be compatible with it
- A recognized framework that does not depend on a local mandate
- GDPR (EU) and CCPA (California) controls built in
- LGPD-compliant consent for Brazil (NR-1 programs)
- Explicit, disclosed monitoring — never covert
- Hosted on Google Cloud infrastructure, SOC 2 Type II certified
- Encrypted in transit (TLS 1.3) and at rest (AES-256)
- Isolated, ephemeral session processing
Frequently Asked Questions
Does Decodeme store employee messages?
Not the messages themselves. Personal identifiers are removed before an analysis is stored, and the original message is discarded after 72 hours — what remains preserves the tone and risk signal, not the identity. Reporting to the organization is aggregated to the team level, so individuals are never exposed, and no searchable archive of employee communication exists.
Is Decodeme legal under the EU AI Act?
Yes for its core text-based model. The EU AI Act's Article 5 prohibition on workplace emotion recognition is limited to biometric data — facial, vocal, physiological. Decodeme infers signals from written text only, which falls outside that prohibition. Note that text-based analysis embedded in consequential HR decisions can fall under the Act's high-risk (Annex III) obligations, which we treat as a separate compliance track.
How is this different from employee surveillance?
Surveillance records and retains everything for later inspection. Decodeme produces an aggregated risk signal from communication that already happens, with consent and team-level anonymization, and does not retain the original content. The goal is to catch psychosocial risk early — not to monitor individuals.
Monitor Communication Risk Without Crossing the Line
Privacy by design, no biometrics, and a legal posture that holds globally — the foundation under every Decodeme deployment.