How Dictaria handles meeting data: audio to answers

Choose how your meeting data is processed

Start with the mode you choose

There is no single “everything is local” or “everything is uploaded” answer for Dictaria. The path depends on whether you are exploring the demo, transcribing as a guest, or saving a meeting in your account. Audio, transcript text, and AI requests have different paths.

The demo contains fictional, prepared meetings. Editing its notes or checking an example action stores those changes in this browser’s local storage. It does not create an account, record a meeting, or call an AI provider. Demo changes are not automatically imported into a real account.

Guest live transcription runs on the device when the local model is supported. No audio is sent to Dictaria in that mode. The live text is temporary, and captured guest audio is discarded when recording stops. The downloaded speech model may remain cached on the device.

Explore the demo guide

Audio becomes transcript text

For a signed-in meeting, local transcription and cloud transcription are separate choices. Local-only transcription does not upload the audio or generate cloud summaries, but Dictaria does save the locally generated transcript text to your account. “Local audio” therefore does not mean “no saved cloud text.”

When you request cloud transcription, audio chunks travel through the Dictaria backend to a contracted transcription service. The current production path processes those chunks in memory. It does not upload the recording to cloud object storage or build a permanent cloud audio library.

Signed-in browser sessions can retain audio locally for playback, recovery, or download. An unfinished recording with recoverable audio is not automatically removed after 24 hours. Sign-out, account changes, clearing site data, or browser storage cleanup can remove local copies, so download audio you need to keep.

Provider handling is a separate part of this path. The transcription service may retain request content for reliability or abuse monitoring for up to 30 days unless a no-retention control applies. An optional whole-recording upload path is described in the privacy notice, but is disabled in the current production configuration.

Read the audio and processing details

A Coach request adds a question and relevant context

Coach responds when you request advice, an explanation, an answer to the latest question, or an answer to your own prompt. Those controls let you choose when to consult it. They do not mean that every word automatically triggers a new coaching request.

The request uses a bounded portion of the recent transcript and, when selected, reference-document context. That context goes to cloud AI for the requested answer. Local speech recognition does not make the Coach request local. Attach only documents that are appropriate for this processing.

Imagine a client asks whether a proposed launch date includes a review period. You can select the relevant brief and ask Coach to explain the issue. Its response can help you frame a follow-up question, but it is still an interpretation: verify the brief and confirm any commitment with the people involved. This example illustrates a workflow, not a guaranteed answer.

Learn the Coach controls

The meeting becomes a record you can work with

A saved meeting can contain its transcript, overview, notes, enhanced notes, chat, and requested Coach history. Cloud summaries and other AI outputs are generated through cloud processing; an editable note and a generated interpretation are not the same thing as the original words.

Use the transcript to check a name, date, or decision before relying on a summary. Edit the notes into a useful working document, keeping tentative proposals separate from agreed actions. A clean recap should make uncertainty easier to see, not erase it.

Requested text processing also has provider retention rules. The privacy notice describes retention of text prompts and responses for up to 55 days where applicable. This is separate from the retention setting for the meeting saved in your Dictaria account.

Review cloud processing and retention

Workspace Chat finds context in saved meetings

To make meeting text searchable, Dictaria derives smaller passages and numerical representations called embeddings. These help retrieve relevant context from the meetings in your account. Embeddings are derived data used for search, not another audio recording.

When you ask Workspace Chat a question, relevant meeting context is used to compose an AI answer. Source links and quotations give you a route back to the supporting discussion. Open those sources when the answer affects a decision; generated text can still misunderstand a qualification or miss relevant context.

An authorized MCP connection gives a compatible AI tool another way to work with your meeting knowledge. It is a separate access choice with its own permission and revocation controls. Consider the connected tool’s data handling as well as Dictaria’s before sharing context with it.

Understand connected AI tools

Retention and deletion finish the flow

Saved meetings and Workspace Chat use a one-year retention default. Account controls offer 30 days, 90 days, one year, or retention until you delete the content. Automatic cleanup runs daily, so a retention deadline is not a promise of deletion at an exact second.

Deleting a meeting removes its related saved chat, Coach history, and derived search data. Workspace Chat conversations have separate delete controls. Provider retention periods and audio you downloaded outside Dictaria are separate from the saved meeting’s lifecycle.

Choose the mode for the task: the fictional demo to explore, guest local transcription for temporary on-device text, or an account when you want saved meetings and cloud assistance. Then review the current privacy notice for the complete processing, retention, and account-control details.

Find privacy and deletion controls