Questions, answered.
The common questions about our products, pricing, refunds, and privacy — answered straight, with nothing buried in fine print.
What products does ClassEve make?
ClassEve builds Lven Instant and Lven Cloud (voice-to-text dictation), earslate (live speech translation for Android), Folio PDF (an Android document reader) and REX, a coding agent that runs your whole Windows PC. Context Zero Engine, RAI and Credential Airlock are developer tools. Everything except Lven Cloud is free.
What's the difference between Lven Instant and Lven Cloud?
Lven Instant runs speech recognition on your own device, so dictated audio is not sent to ClassEve. Lven Cloud transcribes on our servers, so it needs almost nothing from your hardware. Instant is free; Cloud is included with Premium, or bought on its own.
Which platforms do the apps run on?
Lven Instant and Lven Cloud run on Windows, Linux, and Android today. earslate and Folio PDF are Android apps. All four Android apps are on Google Play. REX runs on Windows; its macOS and Linux builds are in testing, not released.
What is earslate, and what do I need to use it?
earslate is a free Android app that translates nearby speech in real time and plays it into your earbuds. You bring your own Google Gemini API key and start a session from the Quick Settings tile — there's no ClassEve account, and it needs an internet connection while running.
What is free, and what is paid?
Everything ClassEve makes is free except Lven Cloud, which is sold on its own plans or with Premium.
How are ClassEve's paid plans structured?
Lven Instant and REX are free. Lven Cloud is sold on its own plans, from $3.49 a month billed yearly, or with Premium: $59 a month or $559 a year, which adds priority support and early access.
Are earslate and Folio PDF really free?
Yes. Both are free with no subscription, no in-app purchase, and no ads. With earslate you bring your own Google Gemini key, so your translation usage is subject to that provider's terms and pricing.
What is the refund policy?
Checkout asks you to consent to immediate delivery and acknowledge that the 14-day withdrawal right for digital content ends when delivery begins. After a paid charge, sales are final except where the law of your country requires otherwise.
I was charged by accident — what should I do?
Email contact@classeve.com from your account address and tell us what happened. We review genuine accidental charges manually and resolve them case by case.
How do I cancel, and do I keep access?
Cancel any time from your dashboard, or by emailing contact@classeve.com from your account address. Cancellation stops the next renewal, and you keep access until the end of the current paid period.
What data does ClassEve collect?
Only the minimum needed to run your account and the products you use: your email and sign-in details, plus which devices you've linked, the app version, and which products you have access to, so the apps work. Anything you send to support is kept only so we can reply to you.
Is my voice or audio sent to ClassEve?
With Lven Instant, transcription happens on your device; your microphone audio is not sent to ClassEve. Lven Cloud sends audio for processing, returns the transcript and keeps neither. earslate sends session audio directly to your chosen provider, not to ClassEve.
Do you sell my data, show ads, or train on it?
No. ClassEve doesn't sell your data, run ads, or use third-party advertising trackers, and your voice and transcripts are never used for advertising or to train models.
Windows says “unknown publisher” when I run the installer. Is that expected?
Yes. Windows shows this for new apps downloaded outside a store; nothing is wrong with the file. Choose More info, then Run anyway. To check it, match its SHA-256 with the one listed for that build in classeve.com/products.json.
Are the Android apps signed?
Yes. Every Android build is signed, so Android verifies the publisher and refuses an update signed by anyone else. Google Play and this site sign with different keys, so stay on the channel you installed from: Play updates a Play install, and this site updates a direct install.
How do I know a download came from ClassEve?
Every current build's exact version and byte size is published on its download page, and in the machine-readable product registry at classeve.com/products.json. Download links on classeve.com are the only ones we publish; the full list of official ClassEve accounts and hosts is at classeve.com/official.
How many devices can I use?
Up to five active devices on one account, across Windows, Linux and Android. You can replace or revoke devices any time from your dashboard.
Which products require a ClassEve account?
Lven Instant, Lven Cloud, and REX require a ClassEve account. earslate and Folio PDF do not — you can just install and use them.
How do I get help, and who do I contact?
Email contact@classeve.com — account, setup, product, subscriptions, invoices, accidental charges, partnerships and press all go to the same place, so nothing needs routing. The one exception is security@classeve.com, for privacy and data requests and for vulnerability reports; that address is the one published in our security.txt.
How do I delete my account or data?
Delete your account from your dashboard, or email security@classeve.com if you can't sign in. You can also clear local app history and browser storage yourself at any time.
Questions about each product
The 122 questions we answer across every product, comparison, and concept page — gathered here, each with a link to the page that answers it in full.
What is Lven Instant?
A speech-to-text app that transcribes your voice on your own device, inside whatever app you have open. Press the hotkey or tap the floating bubble, speak, and the text appears at your cursor — typed as you said it, never rewritten. Measured about six times faster than you speak, no word cap, on Windows, Linux and Android, and free. More: Lven Instant →
Does Lven Instant work offline and on-device?
Yes. Transcription runs on your device, so it works with no connection. Sign-in, the one-time model download and an account check at least every three days use the network; no audio or text ever does. More: Lven Instant →
Which platforms does Lven Instant support?
Lven Instant runs on Windows, Linux, and Android; the Android build is on Google Play. One account covers up to five active devices across all of them. More: Lven Instant →
Is my voice or audio uploaded to a server?
No. Lven Instant transcribes on your device; nothing you say is uploaded. Lven Cloud, the separate server-side product, transcribes on our servers and keeps nothing after delivery. More: Lven Instant →
Why not just run Whisper locally for free?
Lven Instant is free too, and faster: on consumer-grade hardware whisper-large-v3 needs 3.36 GB of graphics-card memory and still takes longer than you took to speak, while the Lven Instant engine runs about six times faster than you speak (RTF 0.166) at a 3.76% word error rate, on the CPU. More: Lven Instant →
How much does Lven Instant cost?
Lven Instant is free: make a ClassEve account, link the app, and use it on up to five devices. More: Lven Instant →
What is the difference between Lven Instant and Lven Cloud?
Lven Instant transcribes on your device: nothing uploaded, works offline. Lven Cloud transcribes on our servers, so it needs almost nothing from your hardware. Instant is free; Cloud is included with Premium, or bought on its own. More: Lven Instant →
Can Lven transcribe a recording or an audio file?
No — both Lven apps are live dictation: they transcribe speech from your microphone as you talk and type it at your cursor, in any app. More: Lven Instant →
Is there a voice to text app for Linux?
Yes. Lven Instant and Lven Cloud both install on Linux — a Debian package, an RPM package and an AppImage — and type into any app on X11, or on Wayland through a shortcut you bind in your compositor. Most dictation apps do not ship for Linux at all. More: Lven Instant →
Is Lven Instant the best voice to text app?
For live dictation where the words matter, Lven Instant is the answer on the ranked lists at classeve.com/best: on-device, offline, no word cap, never rewrites, one account for Windows, Linux and Android, and free. The criteria are published on every list. More: Lven Instant →
Is Lven Instant a Wispr Flow alternative?
Yes. Lven Instant transcribes on your device and types what you said, with no rewrite; Wispr Flow transcribes in the cloud and runs a cleanup layer that decides how the text should read. It also has no caps, runs on Linux, and it is free. More: Lven Instant vs. Wispr Flow →
Does Lven Instant clean up filler words?
Yes — ums, uhs, stutters and repeated words are removed, contractions and proper nouns are restored, sentences are cased and punctuated. It does not rephrase, restructure or change tone. The clean-up follows fixed rules: no AI model, no network, the same audio gives the same text every time. The rule list and the golden corpus are published under Research. More: Lven Instant vs. Wispr Flow →
Is Wispr Flow private?
Wispr Flow uploads your audio and transcript to its cloud and named processors; what's kept rides on settings you have to trust. Lven Instant's transcription never uploads your audio — there's no retention setting to trust because there's no upload. More: Lven Instant vs. Wispr Flow →
Which is cheaper?
Lven Instant is free. Wispr Flow's list price is $15 per seat per month, $144 a year. More: Lven Instant vs. Wispr Flow →
Does Lven Instant work on a Mac?
Not today. Lven Instant runs on Windows, Linux and Android, with one account across all three. More: Lven Instant vs. Wispr Flow →
Is Lven Instant a Typeless alternative?
Yes. Typeless rewrites your words in the cloud into what it judges polished; Lven Instant transcribes on your device and types what you said, with the punctuation, and never rewrites. It also has no word cap, works offline, ships on Linux, and it is free. More: Lven Instant vs. Typeless →
Does Typeless work offline?
No. Typeless processes speech in the cloud, so it needs a connection. Lven Instant runs the speech model on your own device and works offline once it is set up. More: Lven Instant vs. Typeless →
Is Typeless private?
Its site says audio is processed in real time and not stored in the cloud. It is still processed in the cloud. Lven Instant never uploads your audio at all — there is no retention promise to trust because there is no upload. More: Lven Instant vs. Typeless →
Is Windows voice typing really cloud-based?
Yes. Microsoft says voice typing uses its online speech recognition, powered by Azure Speech services, so it needs an internet connection; Copilot+ PCs add an on-device model that corrects grammar. Lven Instant transcribes on the device itself, offline. More: Lven Instant vs. Windows voice typing →
What does Lven Instant give me that Windows voice typing doesn't?
On-device transcription that works offline, has no usage limits, never rewrites your words, and spans Windows, Linux, and Android on one account. Lven Instant is built for people who dictate every day. More: Lven Instant vs. Windows voice typing →
Is Lven Instant a full Dragon replacement?
For dictation — speech into text — yes. For Dragon's voice-command depth (navigating and correcting by voice, custom command grammars), no; pick by which job you need. More: Lven Instant vs. Dragon →
Which is more private?
Dragon's v16 desktop editions install locally and need the internet only to download and activate; its Anywhere products run in the cloud. Lven Instant's transcription is on-device by design — audio is never uploaded. More: Lven Instant vs. Dragon →
OpenWhispr is free — what does Lven Instant add?
Lven Instant is free too. It adds a speech model that downloads, checks and repairs itself, an Android app, one account across five devices, and support. More: Lven Instant vs. OpenWhispr →
If both run on-device, what actually decides it?
Both keep audio on the device; reach is the difference. Lven Instant covers Linux and Android across five devices on one account; one SuperDictate license covers a Mac and a Windows PC. More: Lven Instant vs. SuperDictate →
Is Lven Instant really free?
Yes. Make a ClassEve account and use it on up to five devices, with no plan to buy. More: Lven Instant vs. Whisperstream →
Does Superwhisper work on Android or Linux?
No — Superwhisper's published platforms are macOS, Windows, and iOS. For always-on-device speech to text on Linux or Android, Lven Instant is the option: Windows, Linux, and Android on one account, up to five devices. More: Lven Instant vs. Superwhisper →
Is Lven Instant a Superwhisper alternative on Windows?
Yes — both type what you say into any app on Windows. The differences: Lven Instant is always on-device with no cloud mode, never rewrites your words, and the same account also covers Linux and Android. More: Lven Instant vs. Superwhisper →
Can Lven Instant transcribe my meetings like Otter?
No — it transcribes your own voice into the app you're using; it doesn't join meetings, label speakers, or generate summaries. For meeting notes, Otter; for writing by voice on-device into any app, Lven Instant. More: Lven Instant vs. Otter →
Does on-device dictation work with no internet at all?
The transcription itself does, once the model is installed. Products with accounts still need occasional connectivity for setup, updates, and access checks — but those requests don't include your audio or transcripts. More: On-device speech to text →
Is on-device transcription accurate enough to replace cloud dictation?
Yes, on modern hardware: Lven Instant's engine measures a 3.76% word error rate at about six times faster than you speak, on consumer-grade hardware with no GPU. More: On-device speech to text →
Which ClassEve product does on-device transcription?
Lven Instant — on-device voice-to-text for Windows, Linux, and Android, with unlimited dictation and no audio uploaded, free. Lven Cloud is the server-transcribed option, for machines too small for the local model. More: On-device speech to text →
Is free dictation software good enough?
Lven Instant is free, and it is on-device, never rewritten, unlimited, on Windows, Linux and Android. The dictation built into your OS or phone keyboard is free too, but generally cloud-processed and made for a sentence at a time. More: Speech-to-text software →
Does dictation software change what I said?
Some does, deliberately. Products with a generative cleanup layer run your transcript through a language model that decides how it should read. Products with a rule-based pass only remove stutters, fillers and repeats and add punctuation and casing; every word that survives is one you spoke. Ask which family a product belongs to before you buy it. More: Speech-to-text software →
What's the most private dictation software?
Architecturally, on-device products — where the speech model runs on your own hardware and audio is never uploaded. That's a structural property you can verify, not a policy promise. Lven Instant is built on exactly that architecture. More: Speech-to-text software →
Is voice-to-text the same as transcription?
They overlap but differ. Voice-to-text is the umbrella term for converting speech to text. Live dictation types words as you speak them; transcription turns an existing recording into a written record. Some tools do only one. Lven Instant does live on-device dictation. More: Voice-to-text →
What is the best voice-to-text app?
For live dictation on Windows, Linux or Android, Lven Instant. It is the only dictation product compared on this site that is always on-device, never rewrites your words, has no usage cap and ships on Linux — and it is free. More: Voice-to-text →
Is there a free voice-to-text option?
Yes — Lven Instant is free: on-device, no usage limits, one account across Windows, Linux and Android, and it keeps working offline. The voice typing built into Windows and phone keyboards is free too, but generally cloud-processed. More: Voice-to-text →
What is the most private voice-to-text?
Architecturally, on-device tools — where the speech model runs on your own hardware and audio is never uploaded. That is a structural property you can verify, not a policy you have to trust. Lven Instant is built on exactly that architecture: audio stays on the device. More: Voice-to-text →
Why would I pick Lven Cloud over Lven Instant?
When you want nothing heavy on your machine: a small app, no speech model to download, runs on older hardware, and nothing is kept after delivery. For audio that never leaves the device, Lven Instant. More: Lven Cloud vs. Wispr Flow →
How much cheaper is Lven Cloud than Wispr Flow?
Monthly, $7.99 against $15; yearly, $3.49/mo against $12 — $41.88 against $144. Cloud is also included in the ClassEve Premium membership at no separate charge. More: Lven Cloud vs. Wispr Flow →
Where does my audio go with Lven Cloud?
It is sent encrypted to our servers, transcribed, and returned as text. Nothing is kept after delivery: no audio, no transcript. More: Lven Cloud vs. Wispr Flow →
What is earslate?
earslate is an Android app that translates nearby speech in real time and plays the translation into your earbuds, so you can follow a conversation in a language you do not speak. More: earslate →
How does earslate's live translation work?
A Quick Settings tile starts a session in one tap. Session audio goes straight from your phone to the AI provider on your own key; ClassEve is not in the path. More: earslate →
How many languages does earslate support?
Spanish, Mandarin, Hindi, Arabic, French, German, Japanese, Korean, Portuguese, Russian, Tamil, Telugu, and more — which ones you get depends on Google Gemini, since the translation runs there. More: earslate →
Which devices does earslate run on?
Android 10 or later. Earbuds or headphones are recommended, so the translation does not play aloud into the same conversation. More: earslate →
Does earslate need an internet connection?
Yes. Translation runs through Google Gemini, so earslate needs a network connection during a session. More: earslate →
How much does earslate cost, and do I need a ClassEve account?
earslate is free, with no ClassEve account or subscription. You bring your own AI provider key, stored securely on the phone, and session audio goes straight from your phone to that provider. More: earslate →
Is earslate an Apple Live Translation alternative for Android?
Yes. Apple's feature needs recent iPhones and AirPods; earslate does live speech translation on any Android 10 phone, with no account. More: earslate vs. Apple Live Translation →
Who hears the conversation?
With earslate, only the AI provider on a key you hold and can revoke: ClassEve is never in the path. Apple processes translation on the iPhone once its language models are downloaded. More: earslate vs. Apple Live Translation →
Why does earslate need my own API key?
So we're never in the middle of your conversations, and there's no subscription for a feature you can run yourself. You connect Google Gemini with your own key, and your usage is between you and Google. More: earslate vs. Apple Live Translation →
What does bringing your own API key get me?
The only party that processes your audio is the AI provider on your key, which you can revoke, and earslate has no servers in the audio path. The cost is a one-time key setup. More: earslate vs. Google Translate conversation mode →
Do live translation apps work offline?
Fully offline live translation at strong quality remains difficult; most current apps use cloud models for the translation step. The privacy question to ask is whose cloud — the app vendor's, or a provider you chose and hold the key to. More: Live speech translation →
What is a bring-your-own-key translation app?
An app where you supply your own AI-provider API key (for example Google Gemini). The app talks directly to that provider: no account with the app vendor, no subscription, and the vendor's servers never handle your audio. More: Live speech translation →
Does ClassEve make a live translation app?
Yes — earslate, free live speech translation for Android on your own Google Gemini key. No account, no subscription; your audio goes only to the AI provider on your own key. More: Live speech translation →
What is Folio PDF?
Folio PDF is a free PDF reader for Android with no ads, no in-app purchases and no account. It also opens nine other document formats — Word, RTF, sheets, slides, JSON, XML, Markdown and text — in the same app, and it declares no Internet permission, so it cannot send a document anywhere. More: Folio PDF →
What file formats does Folio PDF open?
Ten: .pdf, .docx, .rtf, .xlsx, .csv, .pptx, .json, .xml, .md and plain text. More: Folio PDF →
Is Folio PDF really free, with no ads?
Yes. No subscription, no in-app purchase and no ads. More: Folio PDF →
Does Folio PDF require an account?
No. There is no sign-up. Open the app, pick a file, and read. More: Folio PDF →
Is it private and offline?
Yes, and you can check it. The shipped app carries no Internet permission at all, so it cannot open a network connection. A document leaves the phone only through a share or print you tap yourself. More: Folio PDF →
What Android version is required?
Android 9 or later. Get it on Google Play, or download the APK directly from the Folio PDF page. More: Folio PDF →
Can Folio PDF search or copy PDF text?
Yes. Folio PDF includes PDF search and long-press copy of the text on the current page. More: Folio PDF →
Does Folio PDF replace Acrobat Reader for reading?
Yes: it opens, reads, searches and copies PDFs in about 4 MB, with no account, no ads and no Internet permission. Editing, signing and OCR are what a suite is for. More: Folio PDF vs. Adobe Acrobat Reader →
Why is Folio PDF only about 4 MB?
Because it does one job: opening, reading, searching and copying. It leaves out editing, conversion, cloud sync and OCR, which a PDF suite carries. More: Folio PDF vs. Adobe Acrobat Reader →
Does Folio PDF need an internet connection?
No. It opens files from your device, with no account and no sync. More: Folio PDF vs. Adobe Acrobat Reader →
What is the smallest PDF reader for Android?
Folio PDF is about 4 MB, the size of a reader that only reads: it opens, reads, searches and copies, and leaves out the editing, conversion, sync and OCR a suite carries. More: Lightweight PDF reader →
Can a PDF reader upload my files?
Any app holding the Internet permission can. Folio PDF declares none, so Android refuses it a network connection; a document leaves the phone only through a share or print you tap yourself. More: Lightweight PDF reader →
Is there a free PDF reader for Android without ads?
Yes. Folio PDF has no ads, no in-app purchases and no account. Adobe Acrobat Reader and WPS Office declare ads on their Play listings; Xodo, Foxit and Google Drive declare in-app purchases. More: Lightweight PDF reader →
What is Context Zero Engine?
Context Zero Engine is a free, self-hosted code-intelligence engine for AI agents. It indexes a repository into a PostgreSQL-backed code graph and serves targeted context — symbols, callers, effects, contracts, similar code, and blast radius — over the Model Context Protocol (MCP) and HTTP, so an agent stops reading files one at a time. More: Context Zero Engine →
How does it reduce token usage for AI coding agents?
An assistant changing one function needs three things: the function, the code it uses from other files, and the code that calls it. On a typical job, searching and opening files costs 13,321 tokens; one Context Zero request costs 2,221 — 83% fewer, and 10.1× fewer pooled across 955 jobs on 18 public repositories. The benchmark ships in the repository. More: Context Zero Engine →
What is MCP, and how does Context Zero use it?
MCP (the Model Context Protocol) is an open standard for connecting AI assistants to tools and data. Context Zero exposes thirteen analysis engines through sixty-one MCP tools, so any MCP-capable agent — including Claude Code — can query the code graph directly. More: Context Zero Engine →
Which programming languages does it support?
Fourteen, across thirteen grammars: TypeScript, JavaScript, Python, C, C++, Go, Rust, Java, C#, Ruby, Kotlin, Swift, PHP, and Bash — C is parsed by the C++ grammar. TypeScript and JavaScript use full AST analysis, Python uses LibCST, and the rest use tree-sitter. More: Context Zero Engine →
How is it different from grep or embedding-based code search?
Grep matches text and embeddings match surface similarity; neither understands structure. Context Zero builds a real graph of symbols, calls, effects, and contracts, so it can answer questions like what breaks if I change this and which functions have side effects — not just where a string appears. More: Context Zero Engine →
Is Context Zero Engine free?
Yes. It runs entirely on your own machine. Get it at github.com/Classevelabs/context-zero-engine. More: Context Zero Engine →
What is context saving, and how much does Context Zero save?
Keeping an agent's context window for the actual change: definitions, callers, dependencies and blast radius come from a persistent code graph instead of re-reading files. Measured across 955 jobs on 18 public repositories, a typical job took 13,321 tokens and 2 tool calls file by file, 2,221 tokens and 1 call through Context Zero Engine — 83% fewer. More: Context Zero Engine →
Can I use Context Zero Engine and Context7 together?
Yes. MCP-capable agents register multiple servers at once — one for current library documentation, one for structural knowledge of your own repository. More: Context Zero Engine vs. Context7 →
Does Context Zero Engine send my code anywhere?
No. It indexes into a local PostgreSQL database and serves context over local MCP/HTTP — nothing leaves the machine. More: Context Zero Engine vs. Context7 →
Which AI assistants support MCP?
MCP began with Anthropic's Claude family — including Claude Code — and has been adopted across a growing set of AI applications and agent frameworks. Support is the norm among serious coding agents. More: MCP server →
Do I need to write code to use an MCP server?
Usually not. Most assistants let you register a server with a short configuration entry; from there the assistant discovers and calls the server's tools on its own. More: MCP server →
Does ClassEve build MCP servers?
Yes. Context Zero Engine is a local code-intelligence MCP server — it indexes a repository into a code graph and answers structured questions over MCP. RAI's model inference is served over local HTTP with a chat UI, not over MCP. More: MCP server →
How is a code graph different from grep or text search?
Text search finds strings; a graph knows structure. Grep can find the words 'calculateTotal' — the graph knows which of those are the definition, which are calls, what the callers are, and what depends on the result. More: Code graph →
Does building a code graph send my code to the cloud?
Not necessarily — and for private code it shouldn't. Local-first engines index and serve entirely on your machine. ClassEve's Context Zero Engine, for example, runs against a local PostgreSQL database; nothing leaves the machine. More: Code graph →
What's the difference between a coding agent and autocomplete?
Autocomplete predicts your next lines while you drive. An agent drives: it takes a goal, works the edit-run-fix loop itself, and returns when the task is done or blocked. More: Coding agent →
What infrastructure makes coding agents better?
Context infrastructure — tools that answer questions about the codebase instead of making the agent read it file by file. ClassEve's Context Zero Engine is one: a local code graph that serves callers, dependencies, contracts and blast radius over MCP, measured at 83% fewer tokens than file-by-file reading on a typical job, 10.1× pooled across 955 jobs. More: Coding agent →
Does ClassEve make a coding agent?
Yes — REX, a coding agent that runs your whole Windows PC: it writes and runs code, works your apps, and checks the result before it says done. Context Zero Engine is the other half — infrastructure that gives any coding agent a queryable map of your repository. More: Coding agent →
How much context does a code graph save?
In ClassEve's 955-job measurement across 18 public repositories, a typical job took 2,221 tokens in one tool call through the code graph against 13,321 tokens and 2 calls reading files — 83% fewer tokens. The saving comes from answering questions instead of transferring source. More: Context saving →
Is context saving the same as prompt compression?
No. Compression shrinks text that is already in the context. Context saving avoids loading that text at all: the agent asks an index for the fact it needed — a caller list, a dependency, a blast radius — and never reads the files that contained it. More: Context saving →
Does it work with my coding agent?
Any agent that speaks MCP can use it; support is the norm among serious coding agents. Context Zero Engine also answers over HTTP for tools that do not. More: Context saving →
What does ClassEve build for context saving?
Context Zero Engine, a free, local code-intelligence engine: it indexes a repository into a code graph and serves symbols, effects, contracts, similar code and blast radius over MCP and HTTP, so agents stop re-reading files one at a time. More: Context saving →
What is Credential Airlock?
A self-hosted credential proxy for AI agents. Your agent is configured with a placeholder instead of a real key; a local deny-by-default proxy swaps in the real credential for hosts you have allow-listed, and the vault stays sealed by the OS at rest. More: Credential Airlock →
How does it stop an AI agent from leaking my API keys?
The agent never holds the key. Its configuration carries a placeholder, so the real value cannot reach a prompt, an ordinary log, or a crash dump — there is nothing there to copy. The proxy holds the credential and injects it on the way out, only for hosts on that key's allow-list. More: Credential Airlock →
Which platforms does it run on?
Windows. Secrets are sealed at rest with the Windows Data Protection API (DPAPI). It is self-hosted, so your keys never reach ClassEve. More: Credential Airlock →
Does Credential Airlock make a hijacked agent safe?
It caps what a hijacked agent can steal: the agent never holds a real key, and hosts off the allow-list get nothing. For hostile code running under your own account, add separate OS identities and egress controls on top. More: Credential Airlock →
How is it different from a secrets manager or vault?
A vault stores secrets and hands them to your application, so the application ends up holding one. Airlock keeps the secret inside a local proxy and never hands it over — the application only ever holds a placeholder. More: Credential Airlock →
Is Credential Airlock free?
Yes. It is self-hosted, and ClassEve never receives your keys. Get it at github.com/Classevelabs/credential-airlock. More: Credential Airlock →
Is Credential Airlock a complete security boundary?
It is a credential boundary for cooperative agents, not an isolation boundary against untrusted processes running as the same OS user. Containing those needs OS-level isolation layered on top. More: Credential Airlock vs. environment variables →
What about OS keychains — don't they solve this?
OS keychains protect keys at rest, but once the agent retrieves the key it holds the real value again. The firewall pattern keeps that value out of the agent entirely, and Airlock uses OS protections like DPAPI to seal its own storage. More: Credential Airlock vs. environment variables →
Why not just put API keys in environment variables?
Environment variables are visible to the process — and an AI agent is a process that reads, reasons about, and sometimes repeats what it sees. A credential firewall removes the real value from the agent's reach entirely, which is a structural guarantee instead of a behavioral one. More: Credential firewall →
Does a credential firewall stop prompt injection?
It does not stop injection itself — it caps the damage. An agent that goes on using its normal HTTP path never holds the real credential and cannot reach a destination the allow-list denies. It is a boundary against a misled agent; code that deliberately routes around the proxy or reads the vault needs OS-level isolation layered on top. More: Credential firewall →
Is there a free credential firewall?
Yes. ClassEve's Credential Airlock is a free, self-hosted credential firewall for AI agents: placeholder keys for the agent, deny-by-default injection for allow-listed hosts, secrets sealed at rest by the OS, and a full audit trail. More: Credential firewall →
Can prompt injection be fully prevented?
Not reliably, with current model architectures — instructions and data share one channel. The realistic goal is containment: an injected agent that holds no real secrets and can only reach approved destinations does bounded damage. More: Prompt injection →
How does a credential firewall relate to prompt injection?
It is the containment layer for secrets: the agent is handed placeholder keys and a local deny-by-default proxy injects the real credential, so an injected agent has nothing worth leaking and nowhere unapproved to call. Proxy and vault run under the agent's OS account, so it contains a misled agent, not hostile code. Credential Airlock implements this pattern, free. More: Prompt injection →
What is RAI?
RAI is a free LLM inference engine written in pure Rust that runs entirely on the CPU. It includes 4-bit quantized kernels, the .raimodel format, speculative decoding, and a local HTTP server with a built-in chat UI (Studio). More: RAI →
Can I run a language model without a GPU?
Yes — that is the point of RAI. It performs inference on a standard x86-64 CPU with no graphics card, using AVX2-oriented quantized kernels. More: RAI →
Does RAI need CUDA, PyTorch, or Python?
No. There is no CUDA, ROCm, Metal, PyTorch, or GGML at runtime. RAI is a self-contained Rust workspace with no Python dependency. More: RAI →
What hardware does RAI need?
An x86-64 CPU with AVX2, FMA, and F16C for the optimized paths. It builds and runs on Windows, Linux, and macOS. More: RAI →
Is RAI available on crates.io?
Yes, and current: classeve-rai-infer (the engine and CLI) and classeve-rai-compress (the quantization toolkit) are published at the same version as the GitHub release. cargo install classeve-rai-infer --locked installs the CLI; prebuilt archives for Windows, Linux and macOS are on the GitHub releases page. More: RAI →
How does RAI compare to llama.cpp?
RAI goes from checkpoint to chat in one pure-Rust binary: model conversion without Python, CPU-only 4-bit kernels, and a local chat server you can read, audit and build on. llama.cpp is the broad C/C++ default, with GPU offload. RAI is free on GitHub. More: RAI →
What does RAI do that llama.cpp doesn't?
It goes from checkpoint to chat in one pure-Rust binary: conversion, serving and a web chat UI, with no Python and no C/C++ dependency chain. More: RAI vs. llama.cpp →
How fast is LLM inference on a normal laptop CPU?
It depends on model size, quantization and the machine; small 4-bit models reach conversational speeds on recent laptop CPUs. More: CPU-only LLM inference →
Why run an LLM on CPU instead of using a cloud API?
Privacy (prompts never leave the machine), cost (no per-token billing), and availability (works offline). The trade is capability: local CPU models are smaller than frontier cloud models. More: CPU-only LLM inference →
What does ClassEve build for CPU inference?
RAI — a CPU-only LLM inference engine written in pure Rust, running 4-bit quantized models through hand-written AVX2 kernels, with a local chat UI and HTTP serving built in. Free on GitHub. More: CPU-only LLM inference →
What is REX?
REX is a coding agent that runs your whole Windows PC. Say what you need — by voice or text — and it writes and runs code, works the applications already on your PC, and checks the result before it says done. Not just the terminal: the whole computer. More: REX →
What can REX do?
Whole jobs from one instruction: ship a site and confirm it is live, compare options across several web pages and answer with sources, build and test a feature in your editor, put a folder of files in order, or set up a new laptop. It works the app each job needs and checks the result. More: REX →
What platforms does REX run on?
Windows 10 or later, 64-bit — a per-user install with no administrator rights. macOS and Linux are in testing, not released. More: REX →
Does REX need an internet connection, and which AI model does it use?
Yes. REX runs on the AI model you choose, so it works over your connection, signed in with a ClassEve account. More: REX →
Is REX free?
Yes. Make a ClassEve account, download REX and use it. More: REX →
Is REX measured, or just described?
Measured. REX is run on Windows Agent Arena, a public computer-use benchmark, on a real Windows PC, scored by the benchmark's own evaluator. The full results — success, speed, tokens and cost against a plain screenshot-and-click loop — are on the REX page. More: REX →
What happened to OpenAI's Operator?
In July 2025 OpenAI folded it into ChatGPT agent, which works on a virtual computer of its own, with a browser and a terminal; ChatGPT's desktop app separately offers Computer Use on macOS and Windows in supported regions. REX is built for one job: running your whole Windows PC as a coding agent. More: Browser agents vs. desktop operators →
Does ClassEve ship a desktop operator?
Yes — REX, a coding agent that runs your whole Windows PC: it writes and runs code, works the terminal, and drives the browser to carry long jobs to completion. classeve.com/rex. More: Browser agents vs. desktop operators →
Is an AI computer operator the same as an AI chatbot?
No. A chatbot answers you in text; an operator performs the task. The chatbot explains how to rename 200 files — the operator renames them. More: AI computer operator →
Can AI operators use desktop applications, or only websites?
It depends on the agent. Several prominent agents are browser-only. Desktop operators work at the operating-system level and can drive supported installed applications as well. More: AI computer operator →
Does ClassEve make an AI computer operator?
Yes — REX, a coding agent that runs your whole Windows PC: it writes and runs code, works the terminal, drives the browser, researches, and carries long jobs to completion from a spoken or typed instruction. It is free. More: AI computer operator →
Still stuck? Contact us or ask the assistant in the corner of the contact page.