Echo Scribe: The Free Mac App That Replaces Your Dictation, Meeting, and Screen Recording Subscriptions
You are paying between $17 and $30 a month for speech-to-text. Super Whisper wants $30. WhisperFlow wants $30. Otter AI wants $17. Granola wants $20. Add a screen recording tool and a meeting transcription service, and your monthly SaaS bill for "voice stuff" alone passes $80 before you blink.
Here's the thing that bothers me: the hardware inside your Mac can already do all of this. It's fast enough, accurate enough, and private enough. What was missing was the software that connects it without a monthly fee.
Echo Scribe is that software. A free, open-source Mac app that gives you voice dictation in any app, automatic meeting transcription with live call guidance, narrated screen recordings, and a searchable local memory of all of it. Without uploading a single byte to the cloud.
Why subscriptions exist in the first place. Cloud transcription costs money because someone else's servers process your audio. Every minute of audio you upload costs the provider compute, storage, and bandwidth. They pass that cost to you as a monthly subscription and tier it by minutes used. That model made sense when on-device speech recognition was too slow or inaccurate to trust.
It doesn't make sense anymore.
The Parakeet V3 model that Echo Scribe uses runs entirely on your Mac's Apple Silicon, transcribes in 25 languages, and loads and unloads from RAM so quickly you barely notice it's there. You're paying a monthly bill for a server you don't need.
This is the part I keep coming back to. The hardware caught up. The software model didn't.
Voice dictation: one hotkey, every app
Press a hotkey (right Control by default, customizable to anything) and speak. What you say lands as clean, polished text wherever your cursor is. ChatGPT, Claude, Cursor, VS Code, Slack, Gmail, Notion, a browser form. Any app that accepts text input. The local model strips filler words, adds punctuation, and even handles spoken commands like "press enter" or "comma."
I've dictated over 6,000 times now and logged more than a quarter of a million words. My average is around 150 dictations a day. That's not a flex. It's a data point: if you're prompting AI more than 50 times a day, voice dictation is the single highest-leverage change you can make.
Why speaking beats typing for AI prompts. When you type a prompt, you naturally shorten it. You drop context because your fingers are tired. You skip the "because" clause. You write a summary of your thinking rather than the thinking itself.
Speaking reverses that. You use full sentences. You add reasoning. You explain the trade-offs you considered. The model gets a richer input, and the output improves.
I didn't predict this when I started building. It surfaced from watching my own usage patterns. The prompts I spoke were consistently better than the ones I typed. Not because I'm a better speaker than writer. Because speaking removes the friction that makes you edit yourself before you've even finished the thought.
The dictation model is Parakeet V3, an open-source model running fully on-device. It handles 25 languages, supports custom vocabulary for tricky words (technical terms, names, company names), and can be tuned with language automation entries if it consistently misspells something you say.
AI Didn't Get Smarter. It Got Hands. covers AI agent harness in more detail.
Smart dictation: the "echo" trigger word
Plain dictation puts your words down as you spoke them. But sometimes you want the words formatted, as an email, a bullet list, a Slack message, a polished paragraph. Echo Scribe's trigger word feature lets you prefix your dictation with an instruction.
Press the hotkey and say: "Echo, format this as an email... hey Jamie quick update, the sprint review moved to Thursday at 3pm, can you update the calendar and double-check the demo build works."
What lands at your cursor isn't a stream-of-consciousness monologue. It's a clean email in your natural tone.
The trigger word ("echo" by default, customizable to "Jarvis" or anything else) signals the local Gemma AI model to step in and reformat your speech. You can add custom templates for any format you use regularly: bug reports, project updates, design briefs, meeting follow-ups. Because the AI model runs locally on your Mac through llama.cpp, there's no API call, no cloud inference, no per-use cost.
This matters more than it sounds. Local AI formatting means the feature stays free. There's no meter running. You don't think twice before using it.
Meeting transcription with live call guidance
Echo Scribe detects when you join a supported meeting app (Zoom, Google Meet, Slack, FaceTime) and records according to the policy you set per app. Always record. Ask first. Never. The meeting audio is captured locally, transcribed by the on-device model, and stored as a searchable transcript with a summary, decisions, action items, and participant tags.
Live call guidance is the feature that changes how you run meetings. Before a call, you set up a template describing what a great call looks like: the questions a discovery call should cover, the points a sprint review should hit, the qualification criteria a sales call needs to address. During the call, Echo Scribe follows the live transcript and checks off covered points in real time. It can suggest the next question based on what hasn't been covered yet. After the call, it generates feedback on communication clarity, leadership presence, emotional signals, and whether the team adhered to the objectives.
You can run two live guidance templates simultaneously. One tracking a sales discovery framework, another monitoring your own communication patterns. Both run locally on your Mac. Nothing leaves your machine. The feedback is available the moment the meeting ends.
I find myself using this for calls I used to zone out on. Not because I'm more disciplined now. Because the template is watching, and I know it's going to tell me what I missed.
Customize the post-meeting summary template to match your workflow. Meeting notes can be automatically exported to a local folder where other tools can pick them up.
Screen recordings for demos and walkthroughs
If you do demos for clients, walk your team through new features, or reproduce bugs for engineering, you already know that scheduling a call to explain something is slower than sending a recording.
Echo Scribe's screen recorder captures your screen with narration, adds auto-zoom that follows your cursor, provides a custom background or camera preview, and cleans up the audio. The recording can be exported as MP4 or GIF, or uploaded directly to Google Drive for a shareable link. A transcript of the narration is generated so the spoken content becomes searchable alongside your dictations and meeting notes.
The video on this page? Recorded using Echo Scribe's own screen recording feature.
One searchable memory with MCP integration
Dictations, notes, tasks, meeting transcripts, and screen recording narrations all land in one local library on your Mac. Every word is indexed. You can search across everything by keyword, or use the built-in chat to ask questions in plain language: "What did we decide about the pricing page in the last meeting?" The answer comes with source citations showing which transcript or note it pulled from.
Two integration paths connect this memory to the rest of your tools:
Folder export. All your captures can be automatically exported to a folder on your Mac. That folder can be picked up by any external tool: a cloud AI system like ChatGPT or Claude Code, a sync service, or a custom script.
MCP server. Echo Scribe runs a built-in MCP (Model Context Protocol) server. Connect it to Claude Code, Codex, or any MCP-compatible AI agent, and those agents gain tools to search your dictations, notes, meeting transcripts, and screen recording metadata. A connected agent can review your prompting patterns across a full day and suggest optimizations, identify recurring questions, or surface workflow improvements.
After a full day of dictation and meetings, you can ask a connected AI: "What am I doing wrong with my prompts? Is there anything I can automate?" The agent searches your entire day's history and gives you grounded, actionable feedback. Not a generic suggestion.
This is the part that keeps surprising me. I built the MCP server as a utility. But the question I'm now asking every evening ("what pattern did I miss today?") has become the most interesting feature. It turns your own work into something you can study.
Privacy: your data stays on your Mac
No account. No sign-up. No cloud backend. Echo Scribe processes everything locally on your Mac using on-device models. Your dictation audio, meeting recordings, screen captures, transcripts, and notes never leave your machine. There's no subscription, no usage cap, no "pro tier" that unlocks privacy. Privacy is the default.
The app needs microphone access for dictation, Accessibility permission for global hotkeys and cursor insertion, and Screen Recording permission for captures. Camera access is optional for video recordings. All permissions are transparent and revocable.
Because the transcription model (Parakeet V3) and the AI model (Gemma through llama.cpp) run locally, the app works fully offline once the models are downloaded. The only features that require internet are Google Drive uploads and updates.
Install and try it
Echo Scribe runs on macOS 14 or later with Apple Silicon. Installation is a single Terminal command. No Homebrew, no dependency setup, no account creation.
curl -fsSL https://raw.githubusercontent.com/desduvauchelle/echo-scribe/main/install.sh | bash
Run the same command again to update. Model downloads require a few gigabytes of disk space. Everything else runs on-device, free, with no limits.
FAQ
What is Echo Scribe?
Echo Scribe is a free, private Mac app for voice dictation, meeting transcription, screen recording, and building a searchable local memory of your work. Everything runs locally on your device. No account, no cloud uploads, no subscription.
Is Echo Scribe really free?
Yes. Echo Scribe is free to download and use. There's no subscription, no usage cap, no paid tier, and no "pro" version. It's funded as an open-source project by its developer.
How accurate is on-device transcription?
Modern on-device speech recognition models on Apple Silicon Macs achieve strong accuracy for clear speech, without sending your audio anywhere. Accuracy depends on microphone quality and background noise, but in normal environments it matches cloud services.
What apps does meeting detection work with?
Echo Scribe detects Zoom, Google Meet, Slack, Teams, and FaceTime calls automatically. You configure per-app policies: Always record, Ask first, or Never.
Can I search across my past meetings and dictations?
Yes. Everything Echo Scribe captures is indexed on-device. You can search by keyword or use the built-in chat to ask questions about your history. The chat answers are grounded in your own transcripts and recordings.
Does Echo Scribe share my data with AI services?
No. Echo Scribe transcribes and processes everything locally. If you choose to connect through MCP or export files to a folder, you're deliberately sharing that content. By default, nothing leaves your Mac.
What Mac hardware is required?
macOS 14 or later with Apple Silicon (M-series). Intel Macs aren't currently supported. Allow several GB for model downloads on first install.
"
Want to talk this through? Book a free 15-minute discovery call