Your client work.
A place on your desktop.
Capture a meeting, review what matters, and bring approved context into the next piece of work. The same Vitale workspace, with microphone and computer audio capture.
Preview 0.1.1 · Mac: macOS 15+, Apple Silicon / Intel · Windows: Windows 11, Intel / AMD x64
Private testing builds. Mac is not Apple notarized; Windows is unsigned and needs Microsoft WebView2. Your computer may block opening these previews. Keep its security protections enabled. Production distribution requires platform signing and real-call acceptance testing.
Capture the conversation
Your voice and meeting audio, started by you.
Choose what carries forward
Edit and approve memories and follow-up tasks.
Keep it with the client
Meeting evidence, preferences, and work in one workspace.
How Vitale uses meeting context
- Choose the client and capture or import. The meeting is stored as internal source material. Recording starts only after you confirm it.
- Review a summary and proposed changes. Vitale compares the transcript with approved client memory, shows supporting excerpts, and identifies existing memories a proposal would replace.
- Approve memory and tasks separately. Client preferences and context become active only after review. Follow-up tasks go into the client plan. A task does not automatically authorize agent execution or messages.
- Carry approved context forward. Memory-enabled Vitale tools retrieve the client’s approved context. Old versions remain in the audit history. Firm-wide guidance requires a separate decision.
The preview stores raw audio on your computer until upload and summary succeed, then removes the local recording. Failed uploads remain in Vitale’s Recordings folder for recovery or manual deletion. Cloud transcripts remain in the client workspace. This preview requires a connection for sign-in, transcription, summaries, and agent work.
Our quality and cost recommendation
Start with OpenAI mini transcription + Gemini 3.8 Flash. About $0.213 per one-hour meeting at gross list rates, or $0.1965 after eligible Google promotional credits through December 2026. The estimate uses 12,000 input and 2,000 billed output tokens; the calculator below includes the assumptions.
We made 14 live provider requests using an authored two-voice fixture, clean and noisy audio, and the same summary inputs. Mini transcription retained all ten checked critical facts in both variants. Gemini 3.8 preserved payment uncertainty and proposed only durable client context. Gemini 2.5 added unsupported payment certainty; Luna produced some excerpts that failed exact matching. This small synthetic trial supports a starting choice, not a real-meeting accuracy guarantee.
For speaker labels, evaluate AssemblyAI Universal-3.5 Pro with diarization and keyterms + Gemini 3.8: about $0.313 per hour before credits. Voxtral Mini Transcribe 2 is a lower-cost challenger. These need integration and testing on the same real recordings. Paying more for transcription did not improve our small fixture enough to justify a default upgrade.
Your first test meeting
- Unzip the download. On Mac, open Vitale.app. On Windows, extract the entire folder and open Vitale.exe; install Microsoft WebView2 Runtime if prompted. Keep all Windows files together.
- Sign in from the desktop app. Complete Google sign-in in your browser, enter the app’s six-digit code, then return to Vitale.
- Choose a test client and record a short call with a consenting tester. Use headphones. Say a corrected dollar amount, an explicitly agreed task, and a preference specific to that client.
- End the recording and check the source transcript, summary, task owner, and supporting excerpts. Edit proposals before approving them. Confirm only approved items appear in the client’s context or plan.
- Repeat with a network interruption after stopping. Recover the saved recording, then check for missing audio or duplicate items. Test your usual microphone and meeting platform before relying on the app.
Automated checks cover Mac audio integrity and recovery, Windows audio/security behavior and actual Windows startup, API authorization and retries, and client-switching controls. Microphone permissions, live system audio, Bluetooth, sleep and long-call behavior still need device testing. Audio is retained locally when an upload or summary fails.
Download the test guide and checksumsMeeting costs, in plain numbers
USD list rates checked September 15, 2026. Capture and transcription are separate from the summary model.
Current combination, estimated
$0.213 / meeting ≈ $21.30 / month
| Engine | Per minute | Your meeting | Details |
|---|---|---|---|
| OpenAI mini transcribe | $0.0030 | $0.180 | Current Vitale integration; economical file transcription. |
| OpenAI 4o transcribe | $0.0060 | $0.360 | Configurable alternative in Vitale; benchmark on client calls. |
| OpenAI GPT-Transcribe | $0.0045 | $0.270 | Tested challenger. No quality advantage on our small noisy fixture; not selected by default. |
| Deepgram Nova-3 | $0.0043 | $0.258 | Alternative; prerecorded, monolingual, pay-as-you-go base rate. |
| AssemblyAI Universal-2 | $0.0025 | $0.150 | Alternative; async base rate. Speaker diarization adds $0.02/hour. |
| AssemblyAI Universal-3.5 Pro | $0.0035 | $0.210 | Alternative; async base rate. Speaker diarization adds $0.02/hour. |
| Mistral Voxtral Mini Transcribe 2 | $0.0030 | $0.180 | Speaker labels and timestamps. Requires a separate integration and evaluation. |
| ElevenLabs Scribe v2 | $0.0037 | $0.220 | Multilingual challenger; not integrated or measured in Vitale. |
| Local Whisper | $0.0000 | $0.000 | No API charge; requires local compute, model download and integration. |
| Model | Input / 1M tokens | Output / 1M tokens | Your meeting |
|---|---|---|---|
| Gemini 3.8 FlashVitale trial default; gross global rate. Eligible 50% credits through Dec 31, 2026 lower effective cost. | $1.50 | $7.50 | $0.0330 |
| Gemini 2.5 FlashTested; added unsupported payment certainty in our fixture | $0.30 | $2.50 | $0.0086 |
| Gemini 2.5 Flash-LiteEconomy alternative | $0.10 | $0.40 | $0.0020 |
| GPT-5.6 LunaTested; some excerpts failed exact evidence matching | $0.20 | $1.20 | $0.0048 |
| GPT-5.6 TerraHigher-cost alternative | $2.00 | $12.00 | $0.0480 |
| Claude Sonnet 5Quality challenger; not measured in this Vitale comparison | $2.00 | $10.00 | $0.0440 |
Current combination: OpenAI mini transcribe + Gemini 3.8 Flash. Assumes 12,000 total input tokens (200 per meeting minute, including context and instructions) and 2,000 billed output tokens including reasoning. Actual speech density, context length and reasoning vary. Gross global synchronous pricing before promotional credits; no cache or batch discount. Excludes additional redaction calls, retries, hosting, storage, taxes, optional recording SDK fees, and subsequent agent work. Only mini and 4o transcription are selectable in this preview; the other engines need integration.
Other capture options
You can import transcripts from an existing recorder. Granola also offers API and MCP integration options. Recall’s desktop SDK is an alternative capture layer for a future integration. These are separate products; this preview uses native Mac and Windows audio capture and does not require either one.