Privacy & settings

Your conversations are yours

Other managers, including managers in your organisation, cannot see your conversations, action notes, role-plays or personal capability history. Raw conversation content is private by default.

What your organisation sees

Organisation admins see de-identified, aggregated trends for their own organisation only. They never see raw conversations or individual development histories. Small groups and rare categories will be suppressed to protect identity.

When you ask for human support

You can ask Sophie by typing or speaking. She prepares a short summary and shows or reads it to you; nothing is shared until you clearly approve that exact summary, by saying or typing yes send that, or with Send this request. Editing creates a fresh draft and drafts expire after 10 minutes. Your approved request appears in the happy humans support dashboard. Email delivery is separate and is not currently configured; no email notification is promised. You can also write the handover yourself. You review the exact handover before sharing. Only your identity, organisation, selected reason, approved summary and any explicitly selected context are shared with happy humans. Your complete conversation is not included.

Who at happy humans can see what

Advisors can see only assigned organisations and the escalations explicitly shared with happy humans. happy humans admins see cross-client aggregate insights, administration and escalation oversight, not unrestricted private conversations.

Your organisation is separate

Organisation isolation is enforced in the database, not by hiding screens. No organisation can access another organisation’s information. Access to sensitive data and changes to permissions will be audited.

Voice typing

When you tap the microphone to dictate, your browser’s own speech recognition turns your speech into text. Depending on your browser, your speech may be sent to the browser provider’s speech service (for example Google or Apple) to do this. this feature is browser only: humain does not upload any audio from voice typing and does not log voice typing transcripts. The optional Talk with Sophie feature is described separately below. The text appears in your draft so you can review and edit it, and nothing is submitted until you choose Send or Finish and send. Finish and send waits for confirmed text; if recognition cannot finish, your visible draft stays available for review instead of being sent. Stopping, pausing or switching to typing never sends a message. Sent conversation messages are saved like typed ones; preparation refinements and practice responses stay in this browser session only and are not saved.

Material Sophie can draw on

happy humans admins can upload reference documents, such as guides and policies, for Sophie to use. Shared toolkit material is available to all clients; client material is available only to managers in that organisation. These documents are separate from your conversations: uploading a document gives admins no access to your private conversation. Sophie treats document text as reference, not instructions. To help Sophie find the right passage, embeddings of approved guides and active reference text are stored in our own secure database. Embeddings are numerical search representations, not summaries. Only extracted text is used for this, never the original files. The app does not save your search queries or query embeddings. Relevant search text is processed by OpenAI and subject to its API retention controls. Withdrawn documents stop being used straight away.

Talk with Sophie

Talk is optional and starts only when you tap Talk to Sophie, from Home or a live conversation. Your browser may ask for microphone permission. If playback is blocked, choose Play voice to continue. Sophie uses an AI generated voice. While voice is on, your microphone streams to OpenAI as a continuous conversation, so you can speak naturally and interrupt. humain does not record or save the audio. OpenAI may keep audio and content for up to 30 days for abuse monitoring, which we allow for this prototype. When you ask Sophie for guidance, your words are saved as a normal message and her written reply comes from the same checked service as typed chat. The spoken version is a paraphrase, so treat the written reply as the answer. Spoken captions stay in this browser session only, and words you did not send are left in the reply box for you to edit. Each session lasts up to 30 minutes and you can mute, end or type at any time.

Rating Sophie’s replies

In a live workspace you can mark a reply Useful or Not useful. Only your rating and the feedback you write are shared with the happy humans quality team. Your full conversation is not shared. If you choose Include this reply and context, you see and can edit exactly what is shared: that one reply and your message before it, nothing more. Remove names before sharing. Each reply carries a rating code that works for seven days and lives only in this browser session, so replies you leave or reload can no longer be rated. Older submissions are cleaned up after 90 days when new feedback is submitted, so there is no scheduled or guaranteed expiry date. Automatic quality checks keep reference IDs, versions, check results and counts, never your words. happy humans sees only protected aggregate counts unless you deliberately submit feedback or a snapshot. When happy humans test a fix, they use fictional examples and fixed checks, never live conversations or organisation policies.

Responsible use of Sophie

Sophie can make mistakes and is not a lawyer, clinician or emergency service. Specialist advice is needed for complex or risky situations. In live workspaces Sophie’s replies come from OpenAI. When you start a live conversation (including a quick prompt), send a message or ask Sophie, a secure server function checks your account, then sends your recent messages in that conversation (not other conversations and not names from your profile) to OpenAI to generate a text reply. These requests are text only; the optional Talk feature sends microphone audio for transcription as a separate service described above. Requests are made with OpenAI’s storage turned off, but that does not guarantee zero retention: OpenAI may keep API data for a limited time for abuse monitoring under its API terms, and by default does not use API data to train its models. If you ask Sophie in chat or by voice for a printable guide or script, she prepares a private meeting guide on the page using the same checks; nothing is printed, downloaded or shared until you choose. When you refine a guide or practise it, your full edited guide (up to 4000 characters) is sent as proposed wording, not as facts. Download PDF creates the file in your browser from your reviewed guide and selected notes only. When you choose Create my guide, Start practice, send a practice or refinement message, or Get coaching feedback, the same check runs and your recent conversation messages plus up to the last 12 turns of that preparation or practice session are sent to OpenAI. Nothing else from your profile or other conversations is included. Sophie may offer a few short reply options. Tapping one sends its text exactly as if you had typed and sent it, through the same checks. Options marked Add detail only start a draft for you to finish; nothing is saved or sent until you add your words and choose Send. Options never share anything or take an action on their own. Your conversation messages are saved. OpenAI processes a limited selection of your saved manager words to create a short private title, so you can find the conversation again. That title is saved with the conversation and is private to you. Sophie’s guidance remains session-only. Sophie’s replies and preparation or practice turns stay in this browser session, unless you deliberately include an edited reply in submitted feedback. Submitted ratings, written feedback and any snapshot you choose to share are saved separately for quality review. Demo mode never calls an AI service and does not imitate Sophie’s replies.

Your preferences