Viora
Viora is a voice AI assistant for macOS only. Hold one shortcut and speak to dictate into any app, rewrite selected text, ask about what is on screen, or hand a small multi-step task to its agent.
What is Viora?
Viora is a voice AI assistant that runs on macOS and nowhere else. Rather than opening an app, you hold a single keyboard shortcut, speak, and the assistant works out what you meant: type, edit, answer or act. Those four modes are the entire product. Write turns speech into finished text inside whatever input field your cursor already sits in, matching the tone of the surrounding context and handling punctuation. Edit takes a passage you have selected and rewrites it from a one-sentence instruction. Ask answers a question about a highlighted passage or a screen region and shows the answer beside the source, with no window switching and no copy-paste. Do hands off a small task, research to gather or a draft to write, and steps through it visibly so you can review each move.
Underneath sits a memory that carries across sessions. Viora keeps the preferences you express, the terms you use and the thread you left unfinished, then reuses them next time you speak. Every stored memory can be viewed, edited or deleted. On that basis the assistant builds a morning brief of the day's to-dos, meetings and follow-ups, raises reminders when the moment arrives, and finishes queued work before you return to the machine.
The app is cloud-assisted rather than local: audio, text and surrounding context travel to Viora's backend, and there is no offline mode. The free plan runs on a fast Haiku model; paid plans use a more capable Sonnet model and fall back to Haiku once the allowance is spent. Seventeen connectors are advertised, from Gmail, Google Calendar, Notion, Slack, Linear and Jira to the native Apple apps. The vendor claims more than 28 languages without publishing a list, and is unusually candid on its own comparison pages that Windows, iOS, Android, a public API and local Whisper models are all absent today. The product is still labelled Early Access, at version 0.8.61.
What it does
- Dictate straight into any macOS app, with tone matched to the context and punctuation handled for you
- Rewrite a passage you have selected from a single spoken instruction about tone, length or stance
- Answer a question about a highlighted passage or a region of the screen, in place, next to the source
- Carry out a delegated multi-step task, visibly, one step at a time and open to your review
- Remember preferences, personal terms and unfinished threads from one session to the next
- Assemble a morning brief of to-dos, meetings and follow-ups, then fire reminders when the moment comes
- Capture live meeting captions and turn spoken discussion into notes
When to use Viora / When not to
A quick filter to help you decide if Viora is the right fit.
When to use Viora
- Developers and AI builders who dictate prompts, review notes and technical documentation full of file names, symbols and acronyms
- Creators turning spoken ideas into scripts, hooks, captions and replies to their audience
- Founders and product managers shaping rough thinking into PRDs, status updates and decision records
- Support and customer success teams writing careful ticket replies and keeping help documentation current
- Office writers who live in Mail, Slack, docs and browsers on a Mac and would rather speak than type every sentence
When not to use Viora
- Windows, iPhone and Android users: the app runs on macOS and nowhere else
- Anyone who needs offline or on-device transcription, since every request is cloud-assisted even in Privacy Mode
- Developers looking for a public API, a bring-your-own LLM key or a local Whisper model, none of which exist today
- Regulated teams that require published HIPAA, SOC 2 Type II or ISO 27001 evidence
- Buyers who need broad language coverage, team admin controls or an identifiable legal entity behind the product
How to use Viora
A typical end-to-end flow, from setup to results.
- Open the download page and pick the build that matches your chip: arm64 for Apple Silicon, x64 for Intel Macs
- Install the DMG on your Mac
- Grant the macOS permissions the app asks for: microphone, accessibility, screen recording, notifications and launch at login
- Create an account with email and password or with Google sign-in, so the desktop app syncs with the web account
- Start the 30-day free trial, which needs no credit card
- Hold the shortcut and speak inside the app you already have open to dictate text
- To rewrite, select the passage first, then say in one sentence what should change
- To ask, highlight a passage or point at a screen region, then put your question
- To delegate, state the intent and let the agent break it into steps you can watch and review
- Fill the personal dictionary with names and jargon, save reusable workflows, and switch on Privacy Mode when a task should leave less behind
Pros & Cons
Pros
- One shortcut covers dictation, rewriting, questions and delegated tasks without leaving the current app
- Works at the Mac system layer, so it is available in browsers, mail, docs, chat tools and code editors alike
- A permanent free plan at $0 with no card, on top of a 30-day free trial
- A clear commitment that voice and text are never used to train any model
- Memory stays under the user's control: every entry can be viewed, edited or deleted
- Privacy Mode reduces what a task leaves behind locally and on the backend
- Rapid release cadence, with thirteen versions shipped between mid-July and mid-August 2026
Cons
- macOS only: no Windows, iOS or Android client, and no mobile companion
- Nothing works offline, and Privacy Mode does not make transcription local
- No public API, no bring-your-own LLM key and no local Whisper model
- No published compliance evidence at all: no HIPAA, SOC 2 or ISO 27001, no DPA, and not a single GDPR mention
- No registered company name, no postal address and no named founder; the only contact address sits on a neighbouring domain
- Language coverage is stated as 28+ without a published list, and no hosting country or region is disclosed
- The vendor names two different model providers on its own pages, Fireworks AI in the legal documents and Google Vertex AI in the pricing FAQ
Pricing & Plans
Viora offers a permanent free plan at $0 with no card required, alongside a 30-day free trial. The lowest paid entry point is the Pro plan at USD 15.00 per month, billed monthly, and the top tier, Max, is USD 100.00 per month. Annual billing is not offered.
- context-aware voice agent
- fast Haiku model
- real-time voice input in any app with optional AI cleanup
- natural voice replies
- long-session context
- personal dictionary
- and connectors such as Gmail
- Slack and Notion
- everything in Free plus the more capable Sonnet model
- long-term memory across sessions
- low-latency voice replies
- proactive reports
- reusable saved workflows
- priority access at busy times and early access to new features
- Viora's top subscription tier
- including all Pro features
Data, GDPR & hosting
A consolidated view of how Viora handles your data.
GDPR overview
There is no GDPR mention anywhere. Searching every collected page and its archived HTML returns zero occurrences of GDPR or General Data Protection Regulation. Viora publishes no trust or security page, no Data Processing Agreement, no formal subprocessor list and no Article 27 EU representative, and it never states a legal basis for processing. What exists is adjacent but unqualified: the Privacy Policy, effective 28 April 2026, says you can request access, correction, deletion, export or restriction of your personal information by contacting the company, subject to identity verification, and it acknowledges that information may be processed in countries other than where you live using appropriate safeguards whose mechanism is never named. Anyone with an EU compliance requirement should treat this as undocumented rather than assume it.
Who owns the data?
The Terms of Service are explicit: you keep ownership of your User Content, which covers audio, transcripts, prompts, selected text, window context, file metadata, notes, feedback, dictionary terms, custom skills and connector data. You grant Viora a limited licence to host, store, transmit, process, display, reproduce and create technical derivatives of that content, but only to deliver, secure, maintain, troubleshoot and improve the service, comply with law and enforce the Terms. Viora states it does not sell personal information and does not use your content to train models. Data is shared with hosting, security, email, authentication and analytics providers, Fireworks AI for model calls, Stripe for payments, Google for sign-in, authorities on legal request, and any successor in a merger or sale of assets.
Reuse rights
Nothing in the Terms requires you to ask permission before reusing what Viora produces from your own material. No output licence is claimed and no commercial restriction is placed on the text, rewrites, summaries or answers you generate. The obligations run the other way. You must hold the rights to whatever you record, transcribe, upload or process, and you must obtain consent from meeting participants wherever law or policy requires it. Viora also disclaims the quality of what it returns: output may be inaccurate, incomplete, delayed or unsuitable, and reviewing it before you publish, send or act on it is your responsibility, particularly for legal, medical, financial, employment or safety decisions. The Acceptable Use policy separately forbids using the service to make automated high-impact decisions without human review. Feedback you submit is a special case: you grant a worldwide, royalty-free licence over the suggestion itself, though ownership of the underlying content does not change hands.
Data retention & training
Hosting summary
Hosting is described in the vaguest possible terms. The homepage says data is securely stored in "our cloud", the Privacy Policy says service data sits in protected backend systems with access controls, and no country, region, cloud provider or data centre is ever named. International processing is acknowledged in one line: information may be processed in countries other than where you live, with appropriate safeguards where required, and no transfer mechanism is specified. The only external processing of desktop task content is stated to be Fireworks AI, with Stripe for payments and Google for sign-in; the pricing FAQ, however, names Google Vertex AI instead, and that contradiction is left unresolved on the site. Local data — recordings, transcripts, history, settings and skills — stays under your macOS user profile. The website itself is served behind Cloudflare on an anycast address, which says where the site is delivered from, not where the data lives.
Things to keep in mind
Risks and trade-offs to weigh before adopting Viora.
- The app needs microphone, accessibility and screen recording permissions at once, which lets it read selected text, the active app, the window title, the browser URL or domain, and the names and paths of selected files
- Meeting mode captures system audio, so other participants are recorded; obtaining their consent is left entirely to you
- Recordings and transcripts are also written to your Mac, and the Privacy Policy warns that anyone with access to your account or your backups may be able to read them
- Privacy Mode is easily overestimated: it reduces persistence, but audio and context still leave the machine for the backend and, when needed, for the model provider
- Output may be inaccurate, incomplete or unsuitable, and the vendor pushes the duty of review back to you, especially for legal, medical, financial, employment or safety decisions
- With no registered company name, no address and no hosting country published, there is no identifiable counterparty if you need to escalate a dispute or force an erasure
- A memory that quietly learns your habits and preferences concentrates a personal profile in one place and builds a dependency that is comfortable rather than deliberate; speaking instead of writing can also erode the effort of formulating your own thinking
Setup & Integrations
Technical difficulty
Setup is consumer-grade. You download a DMG, choosing between the Apple Silicon and Intel builds, and the download page explains how to identify your chip. There is no server configuration, no API key and no deployment step. The only real friction is granting several sensitive macOS permissions — microphone, accessibility, screen recording, notifications and launch at login — which some users will hesitate over. An account is created with email and password or Google sign-in, and each connector needs a one-off OAuth authorisation.
Deployment
Integrations
Behind Viora
Resources
All the official URLs gathered for verification and reference.
Alternatives
Tools that compete with or complement Viora.
Frequently asked questions
Which platforms does Viora support?
How do you actually use it?
Is it just dictation?
Does Viora work offline?
Is my voice used to train AI models?
Can I control what Viora remembers?
What does Viora cost?
Is there an API, and how many languages are supported?
Should you pick Viora?
Viora makes a narrow bet and commits to it. One shortcut, one platform, four modes: write, edit, ask, do. For someone who spends the day typing on a Mac, that is a coherent proposition, and the cross-session memory, the morning brief and the timed reminders push it past simple dictation into something closer to an assistant. The pricing is honest and legible: a genuinely free tier, $15 a month for the capable model and persistent memory, $100 for the top tier, cancellation at any time, and a 30-day trial that asks for no card.
What is missing is proof rather than promise. The privacy commitments read well, and the pledge never to train on customer data appears in both the Privacy Policy and the Terms. But there is not a single GDPR mention on the site, no Data Processing Agreement, no subprocessor page, no certification, and no hosting country named. More striking still, no registered company name, no postal address and no identifiable founder appear anywhere, and the only contact address sits on a neighbouring domain. The vendor even contradicts itself on which model provider it uses, naming Fireworks AI in its legal documents and Google Vertex AI in its pricing FAQ.
The product is young: version 0.8.61, an Early Access label on every page, and a first web archive capture dated 3 August 2026. That youth explains much of the opacity, and the release cadence is genuinely fast. Individuals on a Mac who want their voice to replace the keyboard will find this a capable tool at a fair price. Any team with a compliance function, or anyone needing Windows, mobile, an API or offline processing, should wait.
- Choosing a selection results in a full page refresh.
- Opens in a new window.