It works like ChatGPT, except it runs on your own computer, so your family's conversations stay there, not on a company's servers.
Windows today · Android alpha available now · macOS alpha on request
Ask about health, money, work, anything at all. The model runs right on your PC, and your chats are saved there, not in someone's cloud.
Pick a free on-device model (Gemma, Qwen) and Valence runs it entirely on your PC. It works with the internet unplugged. Small models like 8 GB of RAM; bigger ones want 16 GB and a graphics card. Cloud models run on anything, and even then your history and memory stay on your disk.
Kids Mode is a locked-down chat built for children, with daily time limits and a dashboard so you can see how it’s used.
Deny-by-default: kids can only do what you allow. On-device crisis detection (re-audited at 1,000 fresh turns per model), a 988 crisis guarantee, daily time limits, and a parent dashboard that shows how it's being used.
$59.95, once. One purchase, every platform. No account to create, nothing to renew, and your conversations, memory and models stay yours.
One license, yours for good. The free trial comes first. There is no second purchase when the Mac and Android versions arrive: the same key unlocks them. Conversations and memory are ordinary files on your disk you can browse, and X-Ray shows why the AI answered the way it did, citing the memories it used.
A few seconds of the real thing beats any feature list.
One prompt answered by a local model, then re-asked of Claude in the same thread.
It started as a little overlay for my son.
He'd be deep in something on his computer and have a question, and I wanted him to get an answer in one click, without leaving what he was doing. That was the whole app. Then it grew, each feature scratching my own itch. An AI that remembered our conversations, the freedom to switch between models and providers, controls to curate it all my way: things no single app put in one place.
But the more I learned about how my data was kept and used, even on the plans I paid for, the more it unsettled me. I had become the training material; so would my wife, my kids, my friends. So I went deep on local AI, and found something that surprised me: most everyday things can be done for free, in private, on a model running right on your own machine. Nothing leaving. Nothing harvested.
Then my 10-year-old daughter asked to talk to Gemini, to ask her a question. That stopped me cold. My kids needed somewhere built for them, honest, and clear that this is a tool, not a person with feelings. What I found out there for children was bleak: adult AI in impressionable young hands, and stories of real harm.
The forest in front of my children was vast and dark. I knew I had to be the light that lit their way.
That's where Kids Mode was born. A place I could make safer: where I stay in control, can spot concerning patterns early, and know that when my daughter talks with Botly, our Kids Mode guide, she's having a positive, age-appropriate experience.
I could have kept all this for myself, but most people don't have the time or the know-how, and I never wanted to become another web app sitting on everyone's data. So I made Valence for everyone: honest, private AI you own and run yourself. I'm still building it, and the best ideas keep coming from the people using it. Tell me what's missing, if it fits, I'll add it.
Every other "kid-safe" AI is an adult model in the cloud with a filter bolted on, and asks you to trust it. Kids Mode is built the other way: age-aware guardrails the AI can't be talked out of, locked to the on-device model, with a parent dashboard that hides nothing. Then we attacked it a thousand times to be sure it holds.
It's the same walled-off engine for every profile, not just kids: work, home and code each carry their own model, memory and rules; what your work AI knows, your family's never sees.
Underneath, it does a lot. Run completely local, go pure cloud, use both, switch models mid-thread, memories you own, walled-off profiles, tools, Kids Mode, the list goes on. On the surface it stays calm, and it gets better every release.
On first run, Valence suggests a local model and, with one click, downloads and configures the whole thing for you. No account, no API key, no command line. From then on the model runs entirely on your computer, and your chats are saved on your own disk.
Valence keeps a searchable memory of what matters across every conversation. Hundreds of saved threads, tagged and findable, and you can import your history from ChatGPT, Claude and Gemini so it all lives in one place you own. Edit a memory, delete it, or add your own anytime. And if one keeps surfacing when it shouldn't, thumb it down: Valence learns to leave it out of similar prompts.
Prefer Claude or GPT for a heavy task? Bring your own keys and switch mid-thread, or run the same prompt through two at once and compare. The local model stays free and private; the cloud is there when you want it, on your key, never through us.
Valence isn't the only way to talk to AI, it's the one built to be easy, all in one place, and yours. For people who'd rather use their AI than configure it.
ChatGPT is only OpenAI, keeps your history on their servers, bills monthly, and for kids it's an adult model with a filter on top. Fine, until you want another model, real privacy, or something built for your children.
Valence: all the major providers in one window, history on your disk, and a Kids Mode built for kids: not a filter bolted on.
Free runners and self-hosted front-ends are private, but they're local-only or fiddly to set up, barebones to use, and have no family features or support to fall back on.
Valence: one-click install, local and cloud, memory, profiles, Kids Mode: finished, not a weekend build.
Local agent tools chase autonomy, running tasks on their own, and are notoriously painful to install and wire up. They're a power-user workshop, not an everyday companion.
Valence: a calm companion you open every day, powerful where it counts, obvious from minute one.
Subscriptions are fine for some things, but you pay every month, forever, and the day you stop, the access stops. Valence is bought once. The local AI is free to run; I got tired of subscription-mageddon. Have you?
And that's just one, most people juggle two or three. It's real software you're paying for, but the day you stop, so does the access.
Buy it once. The local AI is free to run. For cloud, bring your own key and pay the provider per use: only for what you actually run, with no flat monthly fee.
Buy Valence, $59.95 →7 days free. The whole app, no card. Buy once and it's yours forever: a key, no renewals, no extra fees. One subscription pays for Valence in three months, then it's free.
We won't pretend the cloud is private, instead we're precise about exactly what leaves, and we're never in the path.
A tool you own: not a service that owns you.
Use the entire app free for 7 days: no card, no account, nothing to cancel. Put it through a week of your real work, then decide. The only way to know if it's for you is to try it.
"I had my father-in-law put Valence through real work. A mountain of hydrogeological survey data to summarize, chart, and turn into a report for his lawyer. He got all of it done in one place, and the results were fantastic."
I'm Chase. I built Valence to fix my own daily frustrations with AI, and I use it every single day. A real human stands behind it, email me and you get me, not a ticket queue.
help@helixailabs.com →Valence is the first tool from Helix AI Labs. My independent studio.
On the fence? Install it. The trial asks for nothing, and if one missing piece is holding you back, tell me what it is. I read every email myself, and more than once the missing piece is what I built next.
Chase · help@helixailabs.com
The trial asks for nothing: no email, no account, no credit card. Download, install, use.
Windows 10/11 (64-bit) · 121.67 MB · free 7 days, then $24.99 for Windows, or $59.95 for every platform. Fetches the graphics support your card needs while it installs.
All-in-one installer (620.18 MB): every graphics backend already in the box, so nothing is fetched while it installs. SmartScreen may warn: More info → Run anyway.
Both installers download your AI model once, on first run. Neither one carries it, so that is the part that always comes over the network. Valence picks a model that fits your machine and sets it up for you. After that it runs with the network unplugged.
Slim: F397B8EABFEBB213B08115434ED8BDA9566E0FBB234EBC33E903BBD321DC367D
All-in-one: 23246F6279FFFF18584847D08BF142435649216C6AD6261308ACD896C0B9C12F
An early alpha, and it behaves like one. The core works, the edges are rough. Email me and I’ll send you the build, usually the same day.
Apple Silicon (M-series) · macOS 14+ · 64.7 MB. Local models run on Apple’s own AI engine (MLX), so it needs an M-series Mac; Intel Macs aren’t supported. First launch needs one trip through System Settings → Privacy & Security → Open Anyway; updates arrive in-app after that.
This is an alpha. The core works, and some features are still being built.
64-bit Android 7.0+ · v1.5.22 alpha · 34 MB. From 1.5.14 on, updates install straight over it. Android will ask you once to allow installs from your browser.
Had an earlier Android build from me? Uninstall it first. This one is signed for official release, so it can’t install over the old one, and uninstalling removes that build’s chats and downloaded model.
Google Play comes later. The iPhone alpha is in App Store review. An All-in-one key covers both; an Android-only key is on the buy page.
Android APK: 09160712807D762348A1DD67514EA94F285C20975FD3AB18713570A4BF7AC473
One key, every platform. Buy the All-in-one key once and it unlocks Valence on Windows, Mac, Android and iPhone — no second purchase when they arrive. Only need one platform? There’s a key for each. Buy a license →
Cloud models run on anything. Local models like 8 GB of RAM or more; the bigger ones want 16 GB and a graphics card. You never have to pick one yourself. On first run Valence looks at your machine, suggests a model that fits, and downloads and sets the whole thing up in a couple of clicks. Curious before you install? Check here.
Runs in your browser, sends nothing anywhere.
Browsers only see a rough sketch of your hardware. Valence does the real check on first run, recommends a model that genuinely fits, then downloads and configures it for you. No account, no API key, no command line.
Valence 1.9.1 · October 19, 2026 · all releases
26 releases since June, all of it built and shipped by one person.