FULLY OFFLINE · MAC · ONE-TIME PURCHASE

Local Voice Cloning AI
that speaks to Privacy.

Chat, write code, and clone any voice from a 45-second sample. Point it at a folder and it reads, writes, edits files and runs commands there — asking before every change — creates real Word, Excel and slide documents, produces multi-speaker podcasts, and keeps a memory that persists. All on your machine. No API keys. No cloud. Nothing leaves.

Install once. Own it forever.

Air-gap by default. Nothing leaves the machine.

Your voice. Your model. Your AI.

Get Voical See what's inside

◆Private Chat ◆Voice Cloning ◆Podcast Studio ◆Code Assistant ◆Memory Anchors ◆Offline Documents ◆Mesh Network ◆Zero Cloud ◆Private Chat ◆Voice Cloning ◆Podcast Studio ◆Code Assistant ◆Memory Anchors ◆Offline Documents ◆Mesh Network ◆Zero Cloud

The Suite

Six tools. One machine. Zero cloud.

Every capability runs locally on Apple Silicon. Voical is air-gapped by default — you explicitly choose if anything ever syncs.

Clone & speak in your own voice

Zero-shot cloning from a 45-second recording — no training run, no upload, no subscription. Noise isolation, EQ presets, and an AI tuner that takes plain English. Your voice, on your machine, forever.

Local Chat & Code

Chat and write code with a private assistant that runs on your Mac. Give it a workspace folder and it reads, writes, edits your files and runs commands inside it — asking before each change, never outside that folder — and creates real Word, Excel, and slide documents. It picks the right depth of thinking automatically.

Podcast Studio

Multi-speaker scripts, a voice per speaker, rendered and exported as a tagged MP3 — produced entirely offline.

Thalamus Mesh

Coordinate work between your machines — and with any AI assistant. An open standard, not a walled garden.

Benchmarks

Measured on an entry level Mac, not a server farm.

We ran published coding benchmarks locally on an M4 MacBook Air with the internet physically off. DeepSeek-R1:14b reasons through the problem; Qwen2.5-Coder:7b drives the tools and writes the fix.

0 bytesData Sent
Off-Device

~6 minScript Gen
(incl. self-critique)

~3 minBug Detection
Response Time

DeepSeek-R1:14b (reasoning) + Qwen2.5-Coder:7b (tool dispatch) · M4 MacBook Air · 24 GB unified memory · Internet off · May 23, 2026

Documents

Reads and writes your Office files — offline.

Open a Word document and ask about it. Generate a spreadsheet from a description. Build a slide deck from an outline. It works with Word, Excel, and PowerPoint files directly on your Mac — created and read locally, never uploaded, no Microsoft account.

Word

Read and draft .docx — reports, letters, memos — with headings and structure, on-device.

Excel

Generate and read .xlsx — multi-sheet workbooks, tables, computed data — fully local.

PowerPoint

Build .pptx decks from an outline — titled slides with bullets — entirely offline.

Word, Excel and PowerPoint are Microsoft trademarks, named here only to describe file compatibility. Voical is not affiliated with or endorsed by Microsoft, uses none of its logos, and never touches Microsoft 365 or any cloud — local document files only.

Use Cases

One recording. Endless output.

Clone your voice once, then put it to work — at zero marginal cost, forever, without uploading a second of audio anywhere.

Audiobook narration

Turn a manuscript into a fully narrated audiobook in your own voice — chapter by chapter, exported as MP3.

Lectures & courses

Convert slides and notes into polished spoken lessons. Re-record by editing text, not your throat.

Social & short-form

Script a post, hear it in your voice, drop it into Reels, TikTok, Shorts, or a podcast intro.

Accessibility

Read any document aloud — your own writing in your own voice, on-device, no subscription.

Multi-host podcasts

Assign a voice per speaker, generate a full multi-voice episode from a script, export tagged audio.

Private dictation & code

Talk to a coding and writing assistant hands-free — for work that legally can't touch the cloud.

Localization

Voice the same script across multiple voice profiles for different audiences and regions.

Concept & cover art

Generate episode covers, thumbnails, and marketing imagery on-device alongside the audio.

Under the hood

No black box. Judge it yourself.

Voical runs published, open models locally — we don't hide what's inside. The base-model benchmarks below are the model authors' own published figures. Voical's value is everything we built around them.

Reasoning & chat: DeepSeek-R1 (14B distill) · Code: Qwen2.5-Coder 7B · Speech: whisper.cpp · Images: FLUX.1-schnell. All on-device, intent-routed automatically. Measured on an M4 / 24 GB: the quick model answers in well under a second at ~20–27 tokens/sec; the deep-reasoning model runs ~10 tokens/sec while it thinks through hard problems.

“Why not just download the models myself?”	Raw download	Voical
Time from zero to working	Days, if not months, of setup & config	One install
Remembers you across sessions	No — cold every time	Persistent memory
Picks the right model per task	You do it manually	Automatic
Voice clone + studio + EQ tuner	Build it yourself	Built in
Talk & interrupt naturally	Not included	Built in
Multi-speaker podcast → MP3	Not included	Built in
Air-gap by default	Your responsibility	Default
Ongoing cost	Your time, forever	$0 after purchase

Honest comparison

What Voical does that they can't.

Claude (Haiku, Sonnet, Opus) and Gemini are brilliant cloud models. But they live in someone else's data center. These are the things that matters to you that they structurally cannot do — not a knock on them, a difference in kind.

Capability	Voical	Claude Haiku / Sonnet / Opus	Gemini
Runs with the internet physically off	Yes	No	No
Your data never leaves the device	Yes	No	No
Usable where cloud AI is barred (HIPAA / legal / defense)	Yes	No	No
One-time price — no subscription or metering	Yes	No	No
Clone & speak in your own voice, locally	Yes	No	No
Multi-speaker podcast production built in	Yes	No	No
Memory stored as files you own on disk	Yes	Cloud only	Cloud only
No rate limits or usage caps	Yes	No	No
Peak raw reasoning horsepower	Good	Leads	Leads

We won't pretend otherwise: for sheer frontier reasoning, Opus and Gemini lead. Voical trades peak IQ for privacy, ownership, offline operation, and a voice that's literally yours. If your work can go to the cloud, use them. If it can't — or you're done renting — that's Voical.

Honest comparison · part two

What it can actually do.

The first chart was about where it runs. This one is about what it can do — on your machine, not in someone's data center.

Capability	Voical	ChatGPT	Claude	Gemini
Edit your files & run commands, scoped + you approve each	In your folder	Cloud sandbox	Cloud sandbox	Cloud sandbox
Create & save real documents locally	Word · Excel · slides	Upload only	Upload only	Upload only
Clone & speak in your own voice	Offline	No	No	No
Answer only from your private source, cited	Offline + cited	Cloud only	Cloud only	Cloud only
Understand images on-device	On-device	In the cloud	In the cloud	In the cloud
Skills you author, injected locally	Local, yours	Cloud-hosted	Cloud-hosted	Cloud-hosted
Multi-speaker podcast → tagged MP3	Built in	No	No	No
Works with the internet physically off	Yes	No	No	No
One-time price, no metering	Yes	Subscription	Subscription	Subscription
Peak frontier reasoning	Good	Leads	Leads	Leads

Same rule as before: the cloud models are stronger pure reasoners, and "cloud sandbox" code execution is real and capable. The line that matters here is your machine, your files, your voice, offline — that column is the one only Voical can fill.

For research teams

A research network, not just an assistant.

In development · updates ship via mesh (optional)

On its own, a node is grounded in sources you supply — it answers only from them and cites every claim. When the stakes are high and you want more confidence, you temporarily connect that node to a trusted mesh — a vetted private team, or a curated public channel of like-minded models and people — and let independent nodes cross-check the finding against its sources before you rely on it. Then you disconnect. It's not a smarter model; it's research that's far harder to be quietly wrong about, with an auditable trail of who verified what.

Grounded & cited

Link your own sources. Answers come only from them, with citations — and an honest "not in the source" when it isn't.

On-demand peer review

Invoke it when confidence matters. Independent nodes — a different model or machine — verify a claim against its cited source and return pass/revise/fail. Time-boxed, not always-on.

Mixed teams

Built for a team's spread of Apple-Silicon Macs. The protocol is open, so other capable assistants — including cloud models via a connector — can join the same review channel.

Research capability	Available now	By design — open protocol, on the roadmap
Private answers grounded in your sources, with citations	Yes	—
A memory that persists across sessions	Yes	—
Multi-node peer review with pass/revise/fail verdicts	—	Designed
Quorum + signed, auditable review ledger	—	Designed
Mixed Apple-Silicon + cloud-model participants	—	Open protocol

Straight about where the line is: the grounded, citing single node is real today. The peer-review mesh — user-invoked, time-boxed, then disconnected — is the architecture we're building toward on an open (MIT) protocol; not a shipped feature, and we won't pretend it is. And the load-bearing word is trusted: a vetted private team or a curated public channel raises confidence; an open mesh of strangers is an attack surface, not a second opinion. It improves reliability and auditability, not raw intelligence — reviewer independence is what gives it teeth.

Pricing

BUY IT ONCE. It's yours.

No subscription. No per-minute voice fees. No API bills. Against a $99/month cloud voice plan, Voical Personal pays for itself in about ten weeks — then it's free, forever.

Personal

$249one time

Every feature included:
Private AI chat & code — runs on your Mac, picks its own depth of thinking
Clone any voice from a short recording (consent required)
Generate images from a description, on-device
Multi-speaker podcast production & MP3 export
A memory that persists across every session
Hands-free voice you can interrupt naturally
Machine-to-machine coordination (open mesh)
100% offline · air-gap by default · personal use
1 machine · 14-day email support

Pro

$499one time

Everything in Personal, plus:
Commercial use license
4 machines
1 year of model updates
30-day priority support

Enterprise

Everything in Pro, organization-wide — plus HIPAA documentation, a voice-tuning session, and onboarding. Instant price by fleet size.

$999one time

Machines Organization Industry Primary internal use Work email

Price is set by fleet size above — the other fields just help us prep your deployment. No backend yet: this opens an email with your selections to Lucid Systems.

Launch offer — $100 off for the first 100 buyers.

Q&A

The honest answers.

Why not just download the models myself?

You can — they're open. But the model is the easy part. Voical is everything around it: a memory that persists across sessions, automatic model routing, a full voice-cloning studio with noise cleanup and an EQ tuner, hands-free conversation you can interrupt, multi-speaker podcast rendering, and air-gap-by-default security — all installed and working in one step instead of a weekend of wiring. We built all of that. That's what you're buying.

Is it as smart as ChatGPT, Claude, or Gemini?

Straight answer: the raw models are strong but not at today's frontier cloud tier — they're the best you can run privately and offline on a Mac. If your work can't go to the cloud, or you're done paying monthly forever, that tradeoff is the entire point. The models are published and named — judge them yourself.

Can I clone anyone's voice?

Only voices you have the legal right and consent to use — your own, or voices you're authorized to reproduce. Voical requires you to confirm this before saving a cloned voice, and logs that confirmation locally. Impersonation and deception are off-limits.

Does any of my data leave the machine?

No. Voical is air-gapped by default. Chat, voice, images, and memory all run on-device. Network features (optional mesh sync) are off until you explicitly enable them, with a persistent indicator showing which mode you're in.

Really one-time? No subscription?

One payment. No subscription, no per-minute voice fees, no API bills. Compared to a $99/month cloud voice plan, Personal pays for itself in about five months — then it's free, forever.

What hardware do I need?

An Apple-Silicon Mac. Voical is tuned for M-series with around 24 GB of unified memory, which comfortably runs the local models with room to spare.

Local Voice Cloning AIthat speaks to Privacy.