Just launched · Apple Silicon · local models

YourprivatespeechworkstationforMac.

Dettivo turns your Mac into a local speech engine. Dictate into any app, capture meetings without a bot, and give agents access through CLI, REST, and MCP. Built for low-latency Apple Silicon workflows, not cloud transcription round-trips.

Buy lifetime license · $45 See how it worksmacOS 14+ · Apple Silicon · one-time purchase

Mail

mail.app, Re: API launch

To: team@example.com

Subject: API launch

Hey, quick thought on sequencing.

Dettivo · idle

Hold⌥␣to dictate

Built to disappear

59 ms median raw insert

No cloud round-trip

Apple Silicon, local models

One app, whole loop

Dictation + meetings

Same surface for agents

CLI · REST · MCP

§ Languages

Hears you in 99 languages,on the Mac you already own.

WhisperKit tiers and multilingual Parakeet v3, all on-device. Auto-detection on supported engines, per-language profiles for the ones you swap between most.

endefresitptnlsvnodafiplcsruuktrelarhehizhjakovithidmsswafcyand 69 more

99 totalWhisper coverageParakeet v3 multilingualAll on-device

§ Field reportsVol. 01 · MMXXVI

What day-one users say,in their own words.

Nº01
I'm dictating strategy notes one hour and capturing CTO calls the next. Dettivo is the one Mac app that does both, and it never sends the audio anywhere I'd have to defend in a compliance review.
Cedric Oeldorf
value creation
Growth Factors
Nº02
Court files do not belong on third-party servers. Dettivo is the first dictation tool I have used where the architecture matches what the law already requires of me.
Perica Grasarevic
lawyer (and judge) CH
Nº03
My Claude Code prompts used to live in a scratchpad. Now they go from voice to the agent in one hotkey, and the agent can grep what I said yesterday through Dettivo's MCP server. Nothing else in the dictation space ships this.
Max Gfeller
engineer CH
Sutro

§ 01Dictation core

Low-latency speech input. Anywhere on your Mac.

Hold a hotkey and talk. Dettivo runs local speech-to-text, applies the right cleanup path, and types into the app you are already using.

Settings > Hotkeys · Polish · Languages

Three modes. Raw keeps every word. Polish runs deterministic dictionary corrections, no LLM, no network, instant clean output. Enhanced rewrites with your chosen local or hosted model. Every behavior is a toggle, Dettivo never hides a setting from you.

Global hotkeys, per language

Push-to-talk, toggle, or hands-free. Independently configurable per language profile with app-level overrides.

Three modes, one click apart

Raw keeps every word. Polish runs deterministic dictionary fixes with no LLM. Enhanced rewrites with your chosen local or hosted model. Swap modes mid-transcript.

Direct insertion, safe fallbacks

Typed straight into the focused app. Secure fields detected, clipboard fallback when insertion fails, five-second undo window.

Polish rules + app profiles

Auto-cleanup of filler, punctuation, contractions, with per-app style overrides. Five built-in presets + your own.

Whisper mode + power-aware mic

Quiet speech gets a gain boost. Different mic profiles on battery vs AC. VoiceProcessing IO when you want it.

Quick Access recall

Pop the last few transcripts with a hotkey. Arrow keys navigate. Return inserts polished, ⌘+Return inserts raw.

§ 02Dictation, deeper

More than transcription. A local speech layer.

The pieces that make speech feel native: Quick Note straight into Obsidian, translation before insertion, and screen-aware context for the apps where you work.

Finder, Vault

~/Obsidian/Inbox

Inbox/

Hold ⌘⇧N to dictate a note straight into Inbox.

No editor context. No insertion. Goes to the folder you picked in Settings › Notes.

Dettivo · idle

Profile: Standalone · adds front-matter · reveal on save

Quick Note → your vault

Pick a folder, your Obsidian vault, your context-layer workspace, wherever. Dictation goes straight there as markdown. No app stealing focus.

Three note profiles

Three profiles. Standalone creates a timestamped file. Inbox appends to Inbox/. Daily appends to Daily/YYYY-MM-DD.md. Front-matter optional.

Front-matter your way

Tags, source app, mode, creation time, all optional, all editable per profile. Your Obsidian vault stays yours.

Translate before insertion

Enable translation in Settings > Polish. Speak in one language, Dettivo inserts in another. Per-language default.

Per-language, per-app

Per-language vocabulary, custom dictionary, and per-app polish profiles. Dettivo learns what you actually say.

Custom dictionary

Settings › Dictionary

Recognition rules

detivoDettivo
mlxMLX
gordanGordon
apiAPI

Per-language. Applied before polish. App-specific profiles can extend it.

Slack

Slack, #team-dach

#team-dach

Type a message…

⌘↵ to send

Dettivo · idle

Dettivo, pipeline

Enhanced · context-on

Raw speech

Context attached

Frontmost appVS Code · PR review
SelectionhandleCheckoutSession(...)
Window screenshotOpted-in · 1 frame
Language profileEnglish · engineering

Enhanced output

App profile → Technical · rules respected

Context-aware Enhanced

Dettivo reads the frontmost app, the selection, and (opt-in) a single window screenshot. Enhanced mode uses it to know what you’re actually talking about.

First-class for IDEs + coding agents

Dictate into VS Code, Cursor, Zed, JetBrains, and coding agents like Claude Code with no extensions to install. Enhanced mode knows you’re writing code and behaves like it.

Learns your vocabulary

Dictionary, app-profile, and language-profile overrides stack. Say a typo-prone word once, teach Dettivo to hear it right every time.

§ 03Meetings

Local meeting capture. Without a bot.

System audio plus microphone. Live transcript. Speaker labels. Editable notes and summaries when you stop recording. No guest joins the call, no cloud transcription service is required.

Dettivo, Meetings

Meetings · Weekly sync

Recording

System audio + mic

00:00

System audio healthyMic healthy

No bot. Local capture.

Reliable system audio

Core Audio tap by default, ScreenCaptureKit fallback. Dettivo warns you before capture degrades, not after.

Live → final, cleanly

Live transcript while recording, final canonical transcript after stop. Speaker labels backfill when diarization runs.

Notes + analysis in one surface

Editable notes live next to AI summary, decisions, action items, and a follow-up draft. Rerun analysis without losing what you wrote.

§ 04Meetings, deeper

Meetings that become local memory. Not a vendor's database.

Every meeting can become local files, searchable memory, and a recap draft. Speaker relabeling remembers who is who next time.

Finder, Meetings vault

~/Obsidian/Meetings

Meeting Mirror

Weekly sync · Platform · 32 min

Recording stopped

./2026-04-21/0900-weekly-sync/

✎
transcript.md
Full raw + polished transcript, speaker-labeled, timestamped.
14 KB
✎
notes.md
Summary, decisions, action items, blockers, follow-up draft.
3.2 KB
✎
details.md
Metadata: attendees, duration, provider, diarization, calendar.
0.6 KB

Security-scoped bookmark · survives restartSettings › History › Meeting Mirror

Mail, Weekly sync · recap

Gmail · Auto-compose

Meeting Recap · draft

Elena WeberMarco RossYou

SubjRecap, Weekly sync · 21 Apr

Summary

API cutover scheduled for 8 May · 09:00 Zurich. Elena owns the final load test, Marco drafts the rollback plan.

Action items

• Elena, load test by 6 May
• Marco, rollback doc by 5 May
• You, stakeholder comms

Delivery → Send via Gmail · recipients from calendar · draft-only is one toggle away

Meeting, Speakers

Talk time · 32:04

Diarized + named

Speaker 0
42%
Speaker 1
31%
Speaker 2
27%

Saved names reused next time. No cloud look-up.

Dettivo, Knowledge

Ask across meetings

You asked

Who owned the rollback plan for the API cutover?

Meeting prep agenda

Prep agenda pulls prior decisions + open action items from previous sessions. Walk in already warmed up.

Ask this meeting

Post-meeting Q&A on any single recording. Ask a question, get a cited answer, copy it out.

Auto-everything (if you want)

Auto-title, auto-summary, auto-action-items, auto-follow-up. Every automation is opt-in, auditable, and cancellable.

§ 05Before the meeting

Calendar-aware, no setup

EventKit-aware. Dettivo spots upcoming meetings and offers to record with the agenda, attendees, and title pre-filled.

Upcoming meeting

Starting in 3 min

Weekly sync, Platform

Elena Weber · Marco Ross · You · 45 min

Agenda

• API cutover decision
• Load test results
• Rollback plan ownership

Prompt before the meeting starts

Dettivo reads your local calendar through EventKit. Your events never leave your Mac.

Seeded from your actual calendar

Title, attendees, and agenda seed the capture. Pick the meeting-app target or capture everything.

Interrupted? Recoverable.

The live transcript checkpoint plus retained audio means an interrupted meeting is not a lost meeting.

§ 06Resilient by design

Nothing gets lost

If system audio drops, Wi-Fi dies, or macOS sleeps mid-call, Dettivo restores the final transcript from the retained audio buffer. No duct tape.

Recover interrupted meeting

Interrupted

Weekly sync · 27 min captured

System audio lost

Provisional transcriptsafe

Retained audio buffer27:14

Live checkpoint00:24:30

§ 07DevX · agents

Voice into your DevX. Agents included.

Dictate code-review comments, commit messages, issue updates, terminal notes, and long agent prompts into the tools already on screen. The same local speech memory is exposed through a bundled CLI, loopback REST adapter, and stdio MCP server.

Shared contract

socket → ~/Library/…/Dettivo/dettivo.sock

rest → 127.0.0.1:45831/v1

mcp → stdio · tools + resources

Peer UID auth · per-user token · 0600 socket · no cloud

~/work · zsh

bundled binary

→ localhost · no cloud · your token

$ dettivo status● running · dictation idle · 2 meetings today$ dettivo transcript latest --format markdown# Weekly sync, Platform
* API cutover 8th, 09:00 Zurich
* Elena: load test$ dettivo mcp config --host claude-desktop --write✓ installed in ~/Library/Application Support/Claude

Every surface is the same IPC. CLI → REST → MCP → Swift app.

CLI · dettivo

Bundled with the app.

status · dictation start/stop · meetings · transcripts · mcp config · polish presets. Script anything, pipe anywhere.

REST · loopback only

One toggle, token-gated.

Settings › API flips on a 127.0.0.1 HTTP adapter over the same IPC. Bearer tokens. Streaming import/export endpoints.

MCP · stdio

dettivo mcp serve

Drop the config line into Claude Desktop, Cursor, or Codex with mcp config --write. Tools: search_transcripts, insert_transcript, start_meeting, more.

Dettivo, Macros

Automations · audit

Heard

“hey dettivo, ship staging”

Deny-by-default · allowlists · every run auditable

, Dictation macros · Settings › Automations

Dictation macros

Map trigger phrases to actions: insert text, run an app command, execute a shell command. Safety policies default to deny. Allowlists are explicit. Every run lands in the audit log.

First-use confirmation

Run once, always allow, or cancel. Dettivo never runs a command it hasn’t been explicitly cleared for.

Audit log

trigger_detected · confirm_shown · run_started · run_completed. Stored locally, exportable, scrollable.

§ 08Works with how you work

Your stack, voice-enabled. No vendor lock-in.

Dettivo sits between your voice, your Mac apps, and your local automation surface. Nothing to install downstream. Nothing to swap.

Ointegrates

Obsidian

Vault as context layer

Quick Note → Inbox/. Mirror Meeting → Meetings/ tree.

✉integrates

Mail · Gmail · Graph · SMTP

Recap + digests

Draft-only, auto-compose, or direct send with your account.

#integrates

Slack · Discord · iMessage

Dictate anywhere

Type into the focused app. Clipboard fallback if insertion fails.

<>integrates

VS Code · Cursor · Zed · JetBrains

First-class for IDEs

Enhanced mode is tuned for code. No extension to install.

※integrates

Claude Desktop · Claude Code · Codex

MCP server, one command away

Your dictations + meetings become tools for your agent.

▣integrates

Apple Calendar

EventKit-aware

Upcoming-meeting prompt seeded with agenda + attendees.

Performance promise

Optimized for first insert.

The hot path is built around one question: how quickly does useful text appear where you are working? Raw dictation lands in the focused app in a few dozen milliseconds. Enhanced adds about 200 ms of local polish, with deterministic cleanup short-circuiting model work whenever it can.

raw stop → insert · median

0ms

stop → final commit

local transcribe finalize

insert into focused app

Median 59 ms · p95 123 ms · Apple Silicon, Accessibility on, model prewarmed. Enhanced adds about 200 ms.

§ 09Quick Access

The last transcript, one hotkey away

A keyboard-first popup for the last thing you said. Search, navigate with arrows, insert polished or raw.

Quick Access

Search recent transcripts…⌘⇧V

Voicemail reply · client onboarding
1h
Hey, just confirming we can kick off next Tuesday. I’ll send a calendar invite.
Quick note · benchmark numbers
2h
API p95 dropped from 180 ms to 42 ms after the caching change. Keep an eye on the 99th.
Slack reply · design feedback
3h
The density feels right. Let’s tighten the inspector padding by 2 px before merging.

↑↓navigate↵insert polished⌘↵insert raw

Search + filter

Filter by mode or source app. Search by content. Jump back to history. Quick Access is for when you forget where you said the thing.

Import audio

Import dictation files into History, or bring an external meeting recording into Meetings so it gets the same local transcript and notes flow.

Full retention control

Per-meeting + global retention. Auto-delete by age. Every artifact path shown in the clear. Delete all is one click.

§ What we chase

Six trade-offs,made out loud.

Six choices Dettivo makes on purpose. The rest of the app follows from them.

Vol. 01Commitments

01
Latency
We chase milliseconds, not benchmarks. The first character on screen matters more than the final transcript.
02
Local by architecture
Privacy is where the audio runs, not a policy page. Speech-to-text stays on Apple Silicon by default.
03
No bots
Your meeting stays between the people in it. We never send a guest into your call.
04
Honest defaults
Review-first corrections. No silent downgrades. Every behavior is a toggle you can see.
05
Open formats
Plain markdown in a folder you control. Transcripts you can grep, agents you can swap, vault you own.
06
Lifetime ownership
Buy once. Three Macs. Signed updates forever. No subscription. No lock-in.

Speech should feelbuilt into the Mac.

Dettivo is not a wrapper around a model. It is a Mac speech workstation that knows which window you are in, which mic you are on, which meeting is about to start, and which folder your notes belong in. The commitment is simple: local by default, fast enough to disappear, complete enough for dictation and meetings, and open enough for agents to use.

Swift · MLX · Core Audio: Native
Raw stop → insert: 59ms
Agent surfaces: 3
Cloud required: 0

§ The math

$45 once. vs. nearly the same monthly forever.

Most tools that do less than Dettivo charge almost what Dettivo costs once, every month. Bar lengths below are proportional to five years of paying. Numbers from public pricing pages.

Dettivo

$45 · once

Wispr Flow

$720 · 5 yr @ $15/mo

Granola

$1,080 · 5 yr @ $18/mo

Otter Pro

$1,020 · 5 yr @ $16.99/mo

Superwhisper

$400 · 5 yr @ $8.49/mo

After three months on any of them, you’ve paid more than Dettivo costs you forever.

See full comparisons Pricing details

§ InsideInside Dettivo

The actual thing.

Enough diagrams. This is the real app, every pixel of the Dettivo you install, captured on a Mac that’s been using it daily.

Dettivo dashboard showing Ready status, Enhanced mode, Active license, and Quick Actions for Start Dictation, View History, Settings, and Automations. — Dashboard · HomeEvery status worth knowing at a glance: session, mode, license, and the three APIs, REST, CLI, MCP, wired and running locally.

Dettivo Meetings workspace: sidebar, meeting list, speaker relabel + talk-time timeline, live transcript, and the AI Analysis panel with summary, prep agenda, action items, and Catch Me Up / Knowledge Search. — Meetings · Full workspaceSystem audio and mic, live transcript, speaker relabel, AI analysis, prep agenda, cross-meeting knowledge search, all in one surface, all running on your Mac.

Dettivo Quick Access popover: searchable recent transcripts with Raw / Polished tags and a preview pane showing mode, source app (Warp), characters, words, and an Insert action. — Quick Access · PopoverHit the hotkey anywhere. Arrow-key through recent transcripts. Return inserts polished, ⌘+Return inserts raw. It closes when you’re done thinking about it.

Settings › PolishChoose the right model for the job. Turn app context, window screenshots, or translation on, per language, per app. Every switch is yours.

Dettivo Settings › Automations pane: dictation macros toggle, app command allowlist, shell command allowlist, shared audit log, meeting recap email automation toggle, automation scope and triggers. — Settings › AutomationsDictation macros, app/shell allowlists, audit log, and meeting recap email automation, all opt-in, all visible, all reversible.

Dettivo History pane: searchable list of recent dictations with timestamps and previews, detail pane showing metadata, raw + polished transcripts, audio playback, and actions for Review Replacement Suggestions and Re-run Transcription. — History · Searchable archiveEvery dictation and meeting, kept locally. Raw and polished side by side. Re-run transcription on any item with a different model.

Dettivo Settings › Hotkeys pane: Push-to-Talk (Option+Space), Toggle Mode (Shift+Space), Re-insert Last, Insert Polished, Insert Enhanced, Capture Note, Quick Access, and Cancel with Escape. — Settings › HotkeysOne hotkey per behavior. Push-to-talk, toggle, re-insert, capture note, Quick Access, rebind anything, record directly into the row.

§ 10Privacy

Private by default

Privacy is architectural, not a policy page. Speech-to-text runs on Apple Silicon by default, meeting capture stays on-device, and the only mandatory network calls are license activation and secure updates.

Local-first

Dictation and meeting capture run locally on Apple Silicon. Hosted note generation is optional and uses transcript text only after local transcription.

No telemetry

No telemetry, no analytics, no background uploads. What you capture stays on your disk.

You control retention

Every meeting has a disclosure acknowledgment, an always-on recording indicator, and a delete path, per-meeting and global.

Most voice apps

Upload your audio to the cloud for transcription. By default, not by accident.
Train their AI on the recordings you make. Some let you opt out; the default is opt-in.
Send a meeting bot into your call, or quietly route the audio through their servers instead.
Bill you every month. Stop paying and the app stops working, history included.

Receipts in /privacy.

Dettivo

On-device end to end. Whisper and Parakeet for speech-to-text, deterministic Polish (no LLM, no network), and our own fine-tuned MLX model for Enhanced, tuned to clean up spoken input, not a generic chat model wrapped in a hotkey. All on Apple Silicon. Cloud STT and hosted-AI providers exist only as opt-in BYOK toggles.
No training, ever. We do not receive your audio or transcripts to train anything. We could not if we wanted to.
System audio captured on-device. No bot in the attendee list, no third party in the audio path.
$45 once. Three Macs. Lifetime updates. No subscription, ever.

By design. Not by toggle.

§ 11Under the hood

What actually runs

The Mac is the platform advantage: Apple frameworks, local models, direct insertion, system audio, and automation surfaces all working together.

Native Swift app

Built on Apple frameworks

Core Audio taps, ScreenCaptureKit, EventKit, MLX Swift. One dedicated Mac app that treats the machine as the speech engine.

Right model for the job

Dictation, meetings, and recall each get a path

Whisper and Parakeet handle speech-to-text. Deterministic Polish handles instant cleanup. Enhanced, summaries, and Q&A use the model that fits the work.

Performance budgets

Measured hot paths, not vibes

Time to first insert, warm starts, model downloads, long-audio throughput, and diarization all get benchmarked because speed is part of the product.

Agent-native surface

Apple Silicon, fully utilized

The same local speech memory is exposed through the app, CLI, REST, and MCP so agents and scripts can use it without a cloud hop.

§ 12License + updates

One license. Three Macs. Forever.

Your license activates Dettivo and signs every secure auto-update. Three active devices per license. Revoke and re-activate anytime.

Dettivo, Settings

Cloud

License + Secure Updates

Activate Dettivo

Your key unlocks Dettivo and signs every auto-update.

DTV-

Keychain + signed updates

License stored in Keychain on first activation. Same license signs every Sparkle update, so you never install a build that wasn’t signed by us.

Three active devices

Switching Macs? Revoke the old device from Settings > Cloud and activate the new one. Three active at a time.

Bare-minimum server state

No cloud account required. We don’t know what you dictate. We only know your license key, its device activations, and your email for recovery.

§ 13FAQ

Questions you probably have

Five quick answers. The full FAQ has everything else.

No. Dettivo is a one-time lifetime license. No recurring charges, ever.

Read the full FAQ

Just launched

One license. Three Macs. Forever yours.

$45 one-time. Secure updates included. Cancel nothing, you paid once.

Buy Dettivo Read the FAQ