# Qurmtessence voice archive

Qurmtessence is voice-first. A signed-in speaker can tap **Talk to QI**, speak a request, and receive a spoken answer. The browser keeps a redacted local transcript so a later visit can continue the conversation.

When the explicit archive preference is enabled, each redacted user/assistant turn is also sent to the authenticated `/api/qi-memory` route. The server stores it in the existing `qi_memory_vault` with a conversation ID, turn index, source (`voice`), request ID, and `audio_stored: false`. The existing semantic recall route can use those records to improve future answers in the same signed-in workspace.

No microphone recording is archived. Audio is processed for transcription and then discarded. Secret-like values are redacted in the browser and again on the server. If the QI session, embedding provider, or Supabase table is unavailable, the UI says so and keeps only the redacted local transcript; it never claims that a durable archive exists.

The **Download redacted archive** control exports the local transcript as JSON. **Clear conversation** removes the local copy only; it does not silently delete server records. Server-side deletion remains an explicit, review-gated operation.

Required server configuration for durable workspace archiving is the same as the existing memory route: `QI_SESSION_SECRET`, `SUPABASE_URL`, `SUPABASE_SERVICE_ROLE_KEY`, and `OPENAI_API_KEY` (plus an optional `QI_EMBEDDING_MODEL`).