Valence doesn't watch you.

The whole promise in one line: Valence collects nothing about you and reports nothing to us. No telemetry, no analytics, no account. Some features do go online, and a few of those are on from the start. Every one of them is listed below, with exactly what it sends and where.

01

What Valence does on its own

Nothing about you, ever. No telemetry. No analytics. No crash reports or logs sent anywhere. No account to create. Valence doesn't track how you use it, what you ask, or who you are.

Two routine things happen without you pressing a button:

Update check At launch, Valence asks our site whether a newer version exists. It's a bare version lookup: no identifier, no usage, nothing about you. It's on by default so fixes reach you, it's one switch in Settings to turn off, and local-only mode blocks it.
Search key check Only if you've added your own Tavily web-search key: when Valence starts, or when you open the web search setting, if the key hasn't been checked in the last 12 hours, Valence quietly checks with Tavily that it still works. It sends the key, nothing else. Right now it runs whenever a key is saved, including in local-only mode and Kids profiles, and that's being fixed.

Your conversations, your memory, and your settings live on your computer as ordinary files you can open, read, back up, or delete. If your machine is offline, Valence still works.

Your license key never phones home.

Your key is verified by cryptography built into Valence itself, and so is the trial: there's no activation call, no server check, nothing to ping. The math proves the key is valid, right on your machine. You could buy Valence, unplug the network, and activate it on a computer that never touches the internet again.

02

When something leaves, and where it goes

Installing an update If you say yes to an update, the installer downloads from GitHub. Valence verifies the installer's cryptographic fingerprint (SHA-256) before it ever runs.
Downloading a local model Only when you click. The model comes straight from Hugging Face, not from us. No usage data, no chats, no patterns reach Helix.
Graphics packs The slim installer fetches the graphics-acceleration pack for your card during setup, and you can fetch one later from Settings. Packs come from our GitHub releases; the request reveals which pack your machine asked for. Every pack is integrity-checked against a fingerprint built into the app before it installs. The all-in-one installer already has every pack in the box.
Using a cloud model Optional. If you connect OpenAI, Anthropic, Google, Azure, xAI, OpenRouter, Vertex AI or Ollama with your own key, each message goes directly to that provider, on your account. That includes your prompt, the conversation so far, any memory, knowledge or web results Valence adds for context, and your attachments. OpenRouter requests also carry a header naming Valence as the app. Each provider has its own privacy policy; read theirs. Helix is never in the middle.
Pictures in replies If a reply links to a picture, it doesn't load by itself. Click it and Valence fetches it from that picture's website, which sees the request. In Kids profiles and local-only mode, linked pictures never load.
Importing old chats Your move, start to finish. Sign in to ChatGPT, Claude or Gemini inside Valence and pull your history out, or export it yourself and load the file. Signing in talks to that company directly. Either way the history lands on your device. Helix never sees a word of it.
Other optional features Each is off until you set it up, and each sends what it needs to the service you chose: HIVE cloud memory (memories and summaries), cloud text-to-speech (the text being read aloud), and image generation (your image prompt).
03

Tools & abilities: what's on from the start

Valence can run tools (MCP) and abilities that reach the web, touch your files, or run code. Here is exactly what a fresh install turns on, and what it doesn't.

On from the start: web Web search, read a web page, and the bundled Web Search tool start on, so the AI can look things up. Searching only works once you add your own Tavily key; then the search words go to Tavily. Without a key, nothing is searched. When it reads a page, it fetches that page from its site. A link you paste into a message is fetched automatically.
Automatic look-ups With a Tavily key and a local model, Valence can look things up for you before answering. That search can use your whole message as the query, so the message goes to Tavily.
On from the start: on-device A few helpers that never touch the network: memory recall, the clock, the calculator, and your own knowledge files.
Off until you say so Every other tool and ability, including any MCP server you add, stays off until you turn it on.
Yours to switch off See what each tool can reach and switch any of them off anytime, including web search. Web tools are off in Kids profiles and in local-only mode.
04

Your files & the diagnostic log

Honest both ways: your chats and memory are saved as plain files, which is why you can open, read, and delete them. They're not encrypted unless you choose to encrypt them.

Encryption at rest Off by default, one switch in Settings. Turn on Data Encryption and new chat and memory files are locked to your Windows account (DPAPI); existing files are re-encrypted the next time they change. An encrypted drive (Windows BitLocker and the like) protects everything else at rest.
Keys & secrets Your AI provider keys and Tavily key are stored encrypted (DPAPI). Tokens you give an MCP server are not yet: they sit in the settings file in plain text, and that's being fixed. The email address inside your license key is also kept in that file.
The diagnostic log Valence keeps a local log, just in case something breaks and you need help. Each time Valence starts, it deletes Valence logs older than 14 days. Two things aren't cleaned up automatically yet: crash logs from the local AI engine and bug-report zips. Delete them by hand from the logs folder (Settings → Diagnostics → Log Files) whenever you like.
It's yours to hand over, or not Logs never leave your machine unless you send them to me. The bug-report button only builds a zip on your computer and opens an email for you to send. I can't reach in and pull anything, by design.
05

What Helix receives

From the app itself, nothing: no chats, no usage, no patterns. There are exactly three ways I ever end up with anything, and all three are you reaching out, not the app reporting in:

When you buy Stripe runs the checkout. They take your name, email, country and card to complete the sale and tax. I never see or store your card. Then my purchase system (on Amazon Web Services, in the US) keeps a purchase record: your email, your license key, which key you bought, the Stripe order number, and when the key was made and emailed. Your email is built into your license key, and it shows up in that system's logs. The key is emailed to you through Resend, an email-delivery service.
When you write to me The contact form takes your name, email, topic and message and passes them through Amazon Web Services and Resend to my inbox. Email, from the form or straight to me, lands in Microsoft 365. I keep it long enough to help you, and I'm the only one reading it.
When you visit the site I look at hit counts, how many people landed on a page, and that's it. My web host keeps standard server logs, as every host does. No cookies following you around, no profile, no pixels, no ad networks. The site's fonts load from Google Fonts, so your browser does contact Google for them.

No Helix account. No Helix cloud. No record of your chats.

The only thing I store about you is the purchase record above, and only if you buy. There's no account, no Helix cloud, and no database of your activity. The privacy isn't a policy I promise to follow, it's how the app is built. If you never buy or write to me, I have nothing about you beyond a web-server log line.

06

Kids Mode, children & COPPA

Nearly every major cloud provider disallows users under 18 under COPPA. So Kids Mode is deliberate about this:

Local models only Kids Mode runs on on-device models only: no frontier or cloud models, no exceptions.
Why that matters Because the model is local, your own computer answers your child. Web tools are off in Kids profiles. Helix collects nothing, so COPPA's data-collection concerns don't apply the way they do to cloud AI. Nothing about your child goes online unless a parent chooses it, as listed below. The search key check in section 01 does still run in Kids profiles today; it sends only the key, and only if one is saved.
Voice typing Your child’s voice never leaves your computer. Valence no longer uses Microsoft’s online speech recognition; it only uses the speech recognition built into Windows, which works on your computer. That isn’t accurate enough yet, so voice typing is off in this version, for everyone.
Pictures in replies Pictures linked in a reply never load in a Kids profile, and they stay unloaded when a parent opens a child’s chat too.
HIVE cloud memory Kids chats are never copied to HIVE cloud memory, even if a parent has HIVE set up and later continues a child’s chat in their own profile. HIVE is switched off entirely while a Kids profile is in use.
A parent continuing a Kids chat If you continue your child’s chat from your own profile with a cloud model, Valence asks you first: use a local model instead, send it anyway, or cancel. Nothing is sent until you choose.
Cloud safety scan: your choice Kids Mode has an optional cloud safety scan. If you turn it on, your child's conversations are sent to your own cloud AI provider to be checked for safety concerns. It's off unless you turn it on, and you can turn it off again anytime.
Parents hold the record Everything Kids Mode keeps so you can review it lives on the device, under your control. Review or delete any of it, anytime. It stays on the machine unless you turn on the cloud safety scan or choose to continue a child’s chat on a cloud model.
Adults & frontier models Grown-ups who want frontier models privately can use pay-as-you-go endpoints like Google Vertex AI, which don't train on your data. Those are not available inside Kids Mode.
07

An honest word on kids' safety

Let's be straight: no guardrail is unbreakable: anyone who tells you otherwise is selling something. Here's what we will say plainly:

Valence's Kids Mode meaningfully raises the floor. Its safeguards significantly increase how safely a child can use AI, and the controls it puts in a parent's hands go well beyond what you'll find almost anywhere else. Nothing is hidden, and nothing is hard to find: you see everything your child does, and every lever is right there for you.

08

Your responsibility

Valence is the workspace, not the AI. The models you run in it are made by other companies or published openly; I don't make them and I don't control what they say. The license agreement you accept when you install says this in full. In plain words:

Answers can be wrong AI can be confidently wrong, out of date, or harmful. Check anything that matters before you rely on it.
Not medical advice Valence is not a doctor, lawyer, financial adviser or therapist, and it isn't meant to diagnose or treat anything. Talk to a qualified professional.
Not for emergencies If you or someone else may be in danger, call 911 or your local emergency number. In the US, call or text 988 for the Suicide & Crisis Lifeline.
What you do with it You're responsible for what you ask, what you do with the answers, and for following the rules of each AI provider you connect.
Kids Mode The parent or guardian supervises. Kids Mode's safeguards reduce risk; they don't remove it.
09

Licensing. The whole deal

Try it 7 days, free

The entire app, no limits, no card, no account. Use everything before you decide.

Buy it $59.95, once

You get a key and it’s yours to keep: a perpetual license for devices you own or control. The All-in-one key unlocks Valence on every platform we ship it on: Windows, Mac, Android and iPhone — no second purchase as each one arrives. Need just one? A Windows key is $24.99 and an Android key is $14.99 (Android is in alpha). Mac and iPhone keys go on sale when those apps are released. Buy a license →

Refunds None, here's why

Your key is verified offline, so once it’s issued there’s no switch I could flip to take it back. A refund would mean you keep the working software and get your money back. That's why the 7-day free trial is your test drive: buy only when you're sure.

Platform Windows desktop

A desktop application for Windows. It runs on your machine, and works fully offline once set up.

Your key is verified offline, so nothing on my side can switch it off remotely, and I never deactivate a key for a customer in good standing. The license ends only on a material breach of the EULA, such as selling or sharing keys. Because an issued key can’t be taken back, I won’t sell through marketplaces that force refunds I can’t honor, not stubbornness, just fairness to a solo developer. Helix AI Labs is a registered business in Washington State, USA.

The legal terms are in the End-User License Agreement, the same one you accept in the installer. This page is a plain-language notice; it isn’t part of that agreement.

Last updated: October 19, 2026 · Questions? help@helixailabs.com