MosMos Review (2026): Pricing, Features & Honest Verdict

MosMos is a free Mac dictation app that turns speech into cleaned-up text in any app and records meetings with speaker labels, summaries and to-dos. Verdict: a sharp idea from a real speech-model company, launched with no privacy policy and a broken download link. Best for: bilingual Chinese and English Mac users. Price: free, no paid plan announced. Rating: 6.0/10

What is MosMos?

We found MosMos at number two on Product Hunt on 18 September 2026 under the tagline “Voice writing that works before, during, and after meetings.” It is a native macOS app with one job split two ways. Press Fn once, talk, press Fn again, and your words land as finished text in whatever field your cursor is in. Press Fn and Shift and it records a meeting instead, then hands back a transcript with speakers and timestamps, a summary and a list of to-dos. The site’s headline is “We speak faster than we type,” and the whole product follows from that sentence.

MosMos comes from a company that builds speech models for a living, which sets it apart from the many dictation apps that wrap someone else’s API. The footer’s Company Website link goes to mosi.cn, the home of MOSI AI in Shanghai. That site’s own description lists its in-house models, MOSS-TTS, MOSS-Transcribe and MOSS-VL, along with a voice creation product and a developer API. The MOSS-TTS family is published openly on GitHub under the OpenMOSS organisation. So the people behind MosMos train the kind of model a dictation app depends on. MosMos requires macOS 15 or later, and the site says Windows and iOS versions are coming soon without giving dates.

What Are MosMos’s Key Features?

It writes differently depending on where you are typing

MosMos reads which app you are in and adjusts the output. The site shows five contexts: chat apps such as Lark and WeChat, Email, Document, Prompts, and Code and Terminal. A spoken “uh, I will give you the first draft by Wednesday, just send me the brand assets first” arrives in a chat window as a short, clean message with the filler removed. The style selector defaults to Auto and can be overridden at any time. This is the right design. Nobody dictates an email the way they dictate a terminal command, and most rivals make you pick a mode by hand.

Three levels of refinement, including none

You choose how much of your original wording survives. Basic outputs exactly what you said with no changes. Standard fixes grammar and sentence flow. Advanced rewrites the whole thing as a concise draft. The example on the site turns a rambling two-sentence thought about testing with a small group into one tight sentence. We like that Basic exists. Plenty of AI dictation tools insist on improving you, and sometimes you need the words you said, in the order you said them, for a quote or a legal note.

A keyword list that fixes names before they go wrong

Add proper nouns, homophones and names to My Keywords and MosMos recognises them first and protects them during refinement. The demo shows “Jordon,” “Nimbis” and “Figmar” corrected to Jordan, Nimbus and Figma. Every dictation user knows this pain. A tool that gets your co-founder’s name wrong ten times a day gets uninstalled, and a personal glossary is the cheapest fix there is. According to the makers on Product Hunt, speech-to-text currently supports Chinese and English only, with more languages promised in future model updates.

Meeting notes with speakers, takeaways and to-dos

Fn and Shift starts a meeting recording. MosMos transcribes continuously, separates the speakers, keeps timestamps, and when the meeting ends it produces key takeaways plus to-do items with assignees and due dates. The sample shows three speakers and three extracted tasks, each tied to a person and a day. That puts a dictation app into the same ring as Granola and the notetaker inside Wispr Flow. It is an ambitious scope for a first release, and it is also the feature with the biggest privacy question attached, which we get to below.

How Much Does MosMos Cost?

MosMos costs nothing today. Every button on the site says “Download Free for Mac,” there is no pricing page (mosmos.io/pricing returns a 404), and the Product Hunt listing is tagged as having free options. We could not find a word about usage caps, a paid tier, or what happens to the price later. For a tool that runs cloud models on your meeting recordings, free forever is unlikely, so treat the current price as a launch condition and not a promise.

For comparison, Wispr Flow has a free plan with weekly usage limits and charges $15 per user per month for Pro, or $12 billed annually. Superwhisper has a free tier with unlimited use of local Whisper models and charges $8.49 a month for Pro. Granola’s meeting notes are free with limited history and $14 per user per month on Business. If MosMos eventually lands anywhere under $10 a month with both dictation and meetings included, it would be the cheapest way to get the pair. Right now nobody outside the company knows.

Where Does Your Audio Go?

This is the section the vendor should have written, and did not. The website footer says “Your history in the macOS app stays on this device.” The only privacy text on mosmos.io is a notice about Google Analytics on the marketing site, and it says in plain words that it “does not cover recordings, transcripts, meeting content, or activity inside the MosMos app.” The obvious addresses for a privacy policy and terms of service both return 404 pages. So an app that listens to your microphone and records your meetings shipped without a published policy covering either activity.

The detail that does exist is in the makers’ replies on Product Hunt. They say ordinary dictation uses local speech recognition. Text polishing sends the text being refined, your instructions and formatting preferences, and a limited amount of context to a cloud model. Meeting mode uploads the recording for speaker-separated transcription and processes the transcript in the cloud. That is a reasonable architecture, and it is roughly what Wispr Flow and Granola do too. But a comment thread is not a policy. It does not say which servers, in which country, for how long, who can access them, or whether your audio trains the company’s models. For a company whose main business is training speech models, that last question is not a small one.

Until a policy appears, we would happily dictate Slack messages and prompts with MosMos and we would not run it in a client call, a board meeting or an interview. If you work under an NDA or in a regulated field, the answer today is no.

Could We Actually Install MosMos?

No, and that shapes this review. On launch day every download button on mosmos.io pointed to mosmos.io/download/mac, and that address returned the site’s 404 page. We tried it several times from two different networks, with the same result each time. It is very likely a deployment slip that gets fixed quickly, and by the time you read this the link may work. But we can only review what we could verify, so everything above describes the product as the vendor presents it, not as we tested it.

A few other gaps point to a team more at home with its domestic market than with a global launch. Community links are a WeChat group, a WeChat official account and RedNote, with X as the only Western channel. We found no support email on the site. None of this makes the software worse, but if something breaks, an English-speaking user has few places to ask.

Who is MosMos Best For?

Use MosMos if you work across Chinese and English on a recent Mac and you write a lot of short text: chat replies, prompts for coding agents, quick emails. That is the exact user it was built by and for, and a speech company’s own models should handle mixed-language dictation better than a generic Whisper wrapper. It also suits anyone who wants to try context-aware dictation without paying $12 to $15 a month, provided the content is not sensitive.

Skip MosMos if you dictate in any language other than Chinese or English, since those are the only two supported. Skip it if you are on Windows, an iPhone or a Mac stuck below macOS 15. Skip the meeting recorder entirely if your calls include client, legal, medical or HR material, at least until there is a written policy on where recordings go. And skip it if you need a vendor you can email when something goes wrong.

Best MosMos Alternatives

Wispr Flow is the market leader and the safest pick. The free plan covers dictation in any app in more than 100 languages on Mac, Windows, iOS and Android, with a dictionary, snippets and limited notetaker access. Pro is $15 per user per month or $12 billed annually and removes the dictation cap. It has published privacy terms and a HIPAA option on higher tiers, which is exactly what MosMos lacks.

Superwhisper is the choice for people who want audio to stay on the machine. The free tier includes voice to text in any app, meeting recording and unlimited use of local Whisper models in more than 100 languages. Pro is $8.49 a month, with yearly and lifetime options, and adds cloud models, your own API keys and file transcription. It runs on Mac, Windows, iOS and Android.

Granola is the better tool if meetings are the main event. It records without a bot, the Basic plan is free with limited history, and Business is $14 per user per month with unlimited notes, integrations and an API. It does not do dictation, so you would pair it with one of the two apps above.

Final Verdict: Is MosMos Worth It?

MosMos has the right ideas in the right order. Context-aware style, a refinement dial that includes “leave my words alone,” a keyword list, and meeting notes on the same hotkey make for a tidier product than most dictation apps manage in their first year. It is free, and it is made by a company that builds speech models instead of renting them. For bilingual Chinese and English users, that pedigree matters.

But the launch was not ready. An app that records your meetings needs a privacy policy before it needs a Product Hunt page, the pricing is a blank, only two languages work, and on launch day the download link led nowhere. We rate MosMos 6.0 out of 10 on what is published, with room to move once we can install it and once the company puts its data handling in writing. Try it for low-stakes dictation when the download works. Keep it away from anything confidential until the paperwork catches up.