Skip to main content

Transcript

Wisp's Transcript feature records your calls locally and turns them into structured, searchable transcripts without sending raw audio to any outside server.

Unlike many AI transcription tools, Wisp doesn't join your call as a bot: the app recognizes when you join a call and asks for permission to record.

Transcripts are powered by Whisper, a local speech-to-text (ASR) engine built into the Wisp app. Only its model (574 MB) is downloaded separately, when you click to download it. All recording and transcription happens on your device: the audio (microphone and system streams) is captured by the Wisp app, processed, then deleted. No audio leaves the machine by default. Transcripts are written to encrypted local storage and stay there.

When you ask Wisp to summarize, search or otherwise work with meeting notes, the encrypted transcript is sent for processing and is decrypted only inside the same Trusted Execution Environment (Intel TDX enclave) that handles all of Wisp's model calls. See Trusted Execution Environments.

Transcripts work when Wisp runs on this computer. With Wisp connected to a server, switch back to This computer to record.

Recording a call​

  1. Open Transcripts, on the right of the window.
  2. Download the model the first time: the app asks before it downloads the 574 MB speech model and shows its progress. If the download fails, retry it from the same place.
  3. Press Record to capture the call you are on. A Recording pill shows while it runs; Pause and Resume stop and continue it without starting a new note.
  4. Press Stop when the call ends. The transcript appears once it is ready, and a notification says Transcript ready.

Meeting detection does step 3 for you. When an app such as Zoom, Discord or a browser starts using the microphone, a small Wisp window says Meeting detected and offers to record. It only offers: nothing records until you press Record. When the call seems over and the recording still runs, it reminds you to stop. Turn detection off with Turn off meeting detection in that window, or with Detect meetings in Settings → Recording.

Working with notes​

Each recording becomes a note with a Summary and the full Transcript.

  • Search transcripts finds notes by their text.
  • Rename a note (new ones are called New note), or delete it.
  • Copy the summary or the transcript, or Download the note as Markdown.
  • Start a chat about this note opens a chat that mentions it. In any chat, type # to mention a recording, and the agent reads it when it needs it.
  • If a summary fails, Retry summary. If a recording fails to transcribe, its Retry tries again.

The life of the audio​

Raw audio is deliberately short-lived:

  1. Capture. The app records two tracks: the microphone (you) and the system audio (everyone else). The two tracks are what tell you apart from the others in the transcript.
  2. Transcribe. Speech recognition runs on your device, as the meeting goes on.
  3. Delete. Once the transcript is saved and every track is transcribed, the audio is deleted. If a track fails to transcribe, its audio stays on your device so that Retry can try again, until a retry succeeds or you delete the recording. Audio is never uploaded or kept after that.
  4. Keep the text only. What remains is the transcript: timestamped segments in an encrypted database on your device, alongside your chats, summaries, memory and the audit log.