Direct answer — Is Amical worth it in 2026? Yes, if you want free, open-source dictation on a Mac or Windows PC and do not mind some setup. Amical is MIT-licensed, transcribes locally with Whisper, and offers unlimited local dictation on its Free plan. The trade-offs: its context-aware formatting is still labelled Alpha and needs a language model you install yourself (or its cloud), telemetry is on by default, and features like meeting transcription are only on the roadmap.
This Amical review looks at a young open-source dictation app that is growing fast. Amical presents itself as “AI-powered speech-to-text” that “auto-formats your dictation for any app”. You hold a key, speak, and the text lands in the app you are using, formatted for an email, a chat or a note.
We checked every claim below against Amical’s website, pricing page, documentation, GitHub README and release list on 3 November 2026. We reviewed the desktop app for macOS and Windows, not the mobile apps. Where Amical does not publish something, we say so. At the end, we compare it with Weesper Neon Flow.
What is Amical?
Amical is an open-source dictation app for macOS and Windows that types your speech into any text field. Its GitHub README calls it an “AI Dictation App - Open Source and Local-first”, powered by Whisper for speech-to-text and “open source LLMs for intelligent processing”.
The code is published under the MIT licence, which lets anyone read, modify and reuse it. The repository was created in May 2025 and had about 1,600 stars in November 2026. The desktop app is built with Electron and TypeScript, and bundles whisper.cpp through a native add-on.
What the README marks as done:
- Context-aware dictation: output adapts to the active app
- Local models: “one click setup of local models in-app”, with offline use
- Hotkeys and voice macros: plus custom workflows
- Floating widget: a small desktop pill to start and stop recording
What is not finished yet:
- Smart voice notes (summaries, tasks, structured notes): in progress
- MCP integration for voice commands that control apps: planned
- Real-time meeting transcription (microphone plus system audio): planned
How much does Amical cost?
Amical’s local dictation is free, with no word limit. The paid plan only matters if you use its cloud models. These are the prices on Amical’s pricing page in November 2026, in US dollars.
| Plan | Price | What it adds |
|---|---|---|
| Free | $0 | Unlimited local dictation, unlimited devices, custom vocabulary, AI formatting, 2,000 words/week of cloud dictation |
| Premium | $10/month, or $84/year billed annually (shown as $7/month) | Unlimited cloud dictation, shared vocabulary, team management |
| Enterprise | On request | Private model hosting, custom models, SSO/SAML, seat-based licences |
Two details matter. First, the 2,000-word weekly cap covers cloud dictation only; local Whisper transcription has no cap. Second, Free plan support is community-only, through Amical’s Discord. Email support starts with Premium.
That model is common for open-source apps with a hosted service: the software is free, and the company earns money on the cloud it runs.
Does Amical work offline?
Amical works offline once you download a local Whisper model. Its documentation says the app “can run entirely on your machine”, with no account required for the local option. Cloud transcription is a separate choice that requires signing in.
During onboarding, you pick between a local model and the cloud model. The local models listed in Settings range in size:
| Whisper model | Download size |
|---|---|
| Tiny | ~78 MB |
| Base | ~148 MB |
| Small | ~488 MB |
| Medium | ~1.5 GB |
| Large v3 Turbo | ~1.5 GB |
| Large v3 | ~3.1 GB |
On macOS, Amical runs Whisper on the Metal backend of Apple Silicon, and the docs say “Intel Macs work, just slower”. Amical does not publish minimum macOS or Windows versions on its quick-start page. The README notes that local Whisper transcription requires macOS 15 or later when building from source; we found no statement on whether the same floor applies to the released app.
What about telemetry?
Amical’s telemetry page says anonymous telemetry is on by default. It covers lifecycle events (launches, recordings started and stopped, errors) and performance data such as transcription latency. You can turn it off with one switch in Settings › Advanced.
Telemetry “never includes” your audio or transcripts, unless you press Report on a history entry, which attaches both to your feedback. Turning telemetry off also disables crash reports and the Report button. For a privacy-first reader, switching it off is the first setting to change.
How does Amical’s context-aware formatting work?
Amical reads which app has focus and formats your speech for it. Its formatting documentation describes four main cases:
- Email: greeting, body and closing on separate lines, professional tone
- Chat (Slack, Discord, iMessage): conversational, no paragraph breaks
- Notes: headings and bullets when the content looks like a list
- Everything else: basic punctuation and casing cleanup
The toggle is marked Alpha, and the docs warn that output can vary between runs. Formatting also needs a language model. Locally, that means installing Ollama (or another OpenAI-compatible server such as LM Studio), pulling a model, and selecting it in Amical. Otherwise, you sign in and use Amical Cloud, which counts against the weekly cloud cap on the Free plan.
This is where Amical asks for the most effort. Raw local dictation takes a single model download. Local formatting adds a second piece of software, several gigabytes of language model, and more memory use while you work. For background on the open speech models behind these apps, see our comparison of Voxtral and Whisper open-source models.
Which platforms does Amical support?
Amical’s desktop app runs on macOS and Windows. It installs from GitHub releases or, on a Mac, with Homebrew (brew install --cask amical). The default push-to-talk key on Mac is Fn, and Fn+Space toggles hands-free dictation.
Linux is not supported. The README says the app “does not yet have a Linux native helper”. Linux users can look at our guide to open-source dictation tools on Linux.
On mobile, the website and README link to App Store and Google Play listings, while the pricing comparison table still marks mobile apps as “soon”. We did not test them. Amical claims support for 100+ languages and mixed-language dictation, though its formatting docs call mixed-language input “an edge case” still being tuned.
On releases: the latest stable version was 1.12.5, published on 25 September 2026. Versions 1.12.7 and 1.13.0-beta.1 were published in October as pre-releases. The project ships often, which is good for fixes, but also means behaviour changes between versions.
Amical vs Weesper Neon Flow: how do they compare?
Both apps transcribe speech on your device, insert text at the cursor and support custom vocabulary. They differ in how much you configure, whether a cloud option exists, and how they are funded.
| Weesper Neon Flow | Amical | |
|---|---|---|
| Platforms | macOS 13+, Windows 10 (19041+) / 11 | macOS and Windows (minimum versions not published); no Linux |
| Licence | Proprietary | MIT open source |
| Cloud transcription | None | Optional, requires sign-in |
| Local formatting | Built in, no extra software | Alpha; needs Ollama or another local LLM server |
| Dictation languages (vendor claims) | 55+ | 100+ |
| Free option | 15-day trial (no credit card), then a Free Plan of 1,500 words per week | Free plan with unlimited local dictation |
| Paid plans | €5/mo, €45/yr, €99 lifetime | Premium for unlimited cloud (see pricing table above) |
| Support | Vendor support and Help Center | Community on the Free plan; email from Premium |
Prices stay in each vendor’s own currency: euros for Weesper, US dollars for Amical. We do not convert them, because a converted price is one neither company charges.
Where Amical is stronger: it costs nothing for local use, its code is open for anyone to audit, and its per-app formatting goes further than plain cleanup once you set up a language model.
Where Weesper Neon Flow is stronger: there is no cloud transcription option to switch on by accident, no language model to install for formatting, and support for macOS 13 Ventura is documented. One honest caveat on our side: transcription never leaves your device, but the Weesper licence is verified online periodically and the app works offline for up to 72 hours between checks.
Want to compare them on your own writing? Start the free 15-day Weesper trial, no credit card required, and dictate the same email in both apps.
How does Amical compare with other open-source dictation apps?
Amical sits in a crowded field of free, local dictation tools. Our Handy review covers a minimal open-source app that also runs on Linux, and our VoiceTypr review looks at another local-first option.
Amical’s distinctive point is context-aware formatting with a local language model. Most other open-source apps stop at raw transcription. The price of that ambition is setup: you choose a Whisper model, and for formatting, you run Ollama alongside it. For a wider view, our list of free offline dictation apps compares the main free options.
Who should choose Amical?
Choose Amical if you are comfortable choosing models, want an open-source app you can inspect, and value formatting that adapts to email, chat and notes. Developers who already run Ollama will find the formatting setup quick.
Look elsewhere if you want dictation that works out of the box without a language model, prefer an app with no cloud option at all, or need dedicated support when something breaks. In that case, download Weesper Neon Flow and try it for 15 days. Setup questions are answered in the Weesper Help Center.
Verdict
In our assessment, Amical is one of the most ambitious open-source dictation apps of 2026. Free unlimited local dictation, an MIT licence and per-app formatting make it easy to recommend for technical users who like to configure their tools.
Its limits are those of a young project: formatting still in Alpha, telemetry on by default, a cloud mode that sits one click away, no Linux build, and roadmap features such as meeting transcription not yet shipped. For a free, open-source option you are willing to tune, it is a strong choice. For packaged offline dictation without extra setup, a finished product fits better.