Dettivo.

• 01

  Subscription is the default

  Wispr Flow at $144/yr for dictation. Otter Pro at $204/yr for meetings. Granola at $216/yr for meeting notes. Five years of any one is $720 to $1,080. Stop paying and the app stops working -- and on the cloud-stored services, your saved transcripts go with it.

• 02

  Your voice in someone else’s cloud

  Wispr Flow says it plainly: "Transcription always happens in the cloud." Otter and Granola both upload meeting audio to AWS. Granola dropped the bot but the recording still leaves your Mac. The audio reaches their servers before you read it back.

• 03

  Bot guest, or just a quieter pipe

  Otter's OtterPilot joins your call as a guest in the attendee list. Granola is bot-free but routes the audio to its transcription provider. Either way, a third party is in the audio path of conversations you might have wanted to keep between you.

• 04

  They train on your speech

  Otter: "We train our proprietary artificial intelligence technology on de-identified audio recordings." Wispr Flow uses dictation data "to improve Flow's features and AI models" unless you turn Privacy Mode on. The polish you get tomorrow was tuned on the words you said today.

• 01

  Speak anywhere

  Push-to-talk or toggle hotkey. Dictate into Slack, Mail, Notion, Google Docs, VS Code, Cursor, Claude Code. Direct keystroke insertion; clipboard fallback when the app refuses; five-second undo.

• 02

  Raw + Enhanced, swap mid-stream

  Raw keeps every word. Enhanced polishes with a bespoke fine-tuned model that runs locally on MLX. No generic chat-model wrapper.

• 03

  Language-aware profiles

  Per-language hotkeys, vocabulary, polish rules, and models. Auto-detect or pin. Custom dictionary applied pre-polish for names and jargon.

• 04

  Secure-field safety

  Detects password fields and refuses to type into them. Pasteboard restored after clipboard fallback. No stray artifacts on cancel.

• 05

  Quick Access recall

  Keyboard-first popup of the last transcripts. Arrow keys navigate, Return inserts polished, Command-Return inserts raw. Filter by mode or source app.

• 06

  First-class for coding agents

  Tuned for VS Code, Cursor, Zed, JetBrains. Dictate directly into Claude Code or Codex CLIs with no extensions. Enhanced knows you are writing code.

• 01

  Core Audio + ScreenCaptureKit

  Core Audio tap primary on macOS 14+, ScreenCaptureKit fallback. Live health warnings before capture degrades, not after. Auto-retry across paths.

• 02

  No bot joins the call

  Zoom, Meet, Teams, anything. Dettivo captures system audio from your Mac. Your counterparty sees nothing in the attendee list.

• 03

  Recover interrupted meetings

  Live transcript checkpoint plus retained audio. An interrupted meeting is not a lost meeting; the provisional transcript survives and the final can be re-run.

• 04

  On-device diarisation

  Speaker separation runs locally. Relabel names in the UI; suggestions are reused next time. Talk-time timeline and per-speaker share built in.

• 05

  Canonical meeting folder

  Every meeting writes to a timestamped folder with transcript.md, notes.md, details.md, and the audio artifact. Works inside Obsidian vaults.

• 06

  Full-text search across history

  SwiftData-backed persistent index. Filter by mode, app, provider; jump back to a moment; replay audio; re-transcribe with a different model.

• 01

  Five presets + custom

  Casual, Formal, Professional, Technical, Creative. Custom presets with your own system prompt, rules, and model override.

• 02

  App-aware context

  Reads the frontmost app bundle ID, your selection via the Accessibility API, optionally a single-frame screenshot. Context badge (Ready, Limited, Blocked) is always explicit.

• 03

  Per-app profiles

  Slack tone different from Mail tone different from Git commit message. Override preset and rules per frontmost app.

• 04

  Local by default, BYO cloud

  MLX polish runs locally on Apple Silicon. Optional OpenAI-compatible endpoint (OpenAI, Ollama, vLLM, LM Studio) for users who want it.

• 01

  Summary + actions + decisions

  AI-generated summary, owner-tagged action items, decisions, blockers, and a ready-to-send follow-up draft. Every section editable in-place.

• 02

  Analysis without stomping notes

  Rerun analysis when the transcript updates. Your edits stay. Copy and export precedence: live editor, then persisted notes, then generated analysis.

• 03

  Post-meeting Q&A

  Ask this one meeting a question. Or ask across every meeting in history, with inline [S1], [S2] citations that jump back to the source transcript.

• 04

  Agenda templates + prep

  Standard, Project Sync, Client/Sales, Standup. Prep agenda pulls prior-meeting highlights and open action items. Timer overlay for 15/30/45/60 minute blocks.

CLI

dettivo status · dictation · meeting · transcripts · import · export

REST

127.0.0.1:45831/v1, bearer auth, streaming import/export

MCP

dettivo mcp serve (stdio, MCP 1.0). Claude Desktop, Cursor, Codex configs

IPC transport

Unix socket, mode 0600, peer-UID auth, hardened token mode

Output formats

plain · markdown · json (SRT/VTT on export)

Surface coverage

Everything the app can do, CLI and MCP can too. Single contract

100% localhost. No public API to wait on, no vendor cloud dependency on the developer surface. Your scripts read transcripts, your agents start meetings, your team never asks whether a third party has the recording.

• 01

  Dictation macros

  Trigger phrases map to actions: insert text, run an app command, execute a shell command. Always-confirm, first-use-confirm, or opt-in auto-allow. Deny by default.

• 02

  Meeting recap automation

  On analysis complete, draft a recap. Three delivery modes: draft-only (nothing leaves your Mac), Mail.app auto-compose (no send), or OAuth send via Gmail or Microsoft Graph.

• 03

  Weekly digest

  Scheduled rollup of open action items. Same delivery modes. Credentials in Keychain, revocable. Idempotent triggers, no double-sends.

• 04

  Calendar-aware prep

  EventKit reads your local calendar. Events never leave your Mac. Upcoming-meeting prompt card with agenda seeded from the event.

• 01

  Local by default

  Audio, transcripts, polish, and analysis land on your disk. Licence activation and Sparkle update checks are the only mandatory network calls.

• 02

  Everything via MLX

  Local inference runs on MLX to use the Neural Engine and unified memory. No Python subprocess, no ONNX shim, no Electron.

• 03

  Bespoke polish model

  The default polish model is fine-tuned for cleaning spoken input, not a general-purpose chat model trying to guess what you meant.

• 04

  Full storage transparency

  Every artifact path shown in Settings. Audio retention per session. Day-based auto-delete policies. No telemetry, no analytics, no background uploads.

Special launch offer (first 500)

USD 29 · one-time · lifetime

Standard

USD 45 · one-time · lifetime

Active Macs

3 per licence · revoke any time

Updates

Lifetime · Sparkle-signed

Refund window

14 days · licence revoked on refund

Subscription

None · ever

Most of the dictation category is subscription. Dettivo is not. The special launch offer is USD 29 for the first 500 buyers; standard is USD 45. Either way, lifetime. The same licence signs every Sparkle update -- you never install a build that was not signed by me.

Platform

macOS 14 Sonoma or newer

Architecture

Apple Silicon (M1 and newer)

Frameworks

Swift 6 · SwiftUI · AppKit · SwiftData

STT engines

WhisperKit, Parakeet; optional cloud STT (BYOK)

LLM runtime

MLX local; optional OpenAI-compatible endpoint

Audio

Core Audio tap primary · ScreenCaptureKit fallback

Updates

Sparkle 2.8 · signed artifacts

Agent surface

CLI · loopback REST · MCP (stdio)

• 01

  Made for Omarchy

  Dettivo sits in the Omarchy bar, follows the active theme on every screen, and uses the F9 keys Omarchy already binds for dictation. It also runs on plain Hyprland, Sway, Niri, and GNOME or KDE through the GlobalShortcuts portal.

• 02

  Hold F9, talk, let go

  The text lands in the window that had focus when you pressed the key, typed through the Wayland virtual keyboard or pasted where the window refuses it. Raw, Polish and Enhanced modes, with a local language model doing the rewrite.

• 03

  Meetings, still no bot

  Your microphone and the call audio record as two tracks on one clock and transcribe live. At stop you get the transcript with speaker names, a summary, decisions and action items. After a crash or a reboot, the meeting is still there to recover.

• 04

  Agents can drive it

  One command registers Dettivo as an MCP server in Claude Code, Codex, Cursor or Claude Desktop, with nineteen tools for searching transcripts, inserting text and starting a dictation. The dettivo CLI speaks --json.

Click to enlarge

Click to enlarge

Install

# Arch or Omarchy: download dettivo-bin-*.pkg.tar.zst
# from the latest release (link below), then:
sudo pacman -U dettivo-bin-*.pkg.tar.zst
systemctl --user enable --now dettivod.socket
dettivo setup omarchy   # keys, bar plugin, recording pill
dettivo doctor          # what is installed, what is missing

Each release carries the pacman package with its checksum. On plain Hyprland, Sway or Niri, run dettivo setup hyprland (or sway, niri) in place of the Omarchy line. A GPU is optional; without one the engines run on the CPU with smaller default models.

dettivo.com/linux →Source on GitHub →v0.2.0 release notes →

Click to enlarge