Your chats stay on your machine. We never see your prompts.
Sixteen subscriptions, or one byself: every major model in one app, and you pay only for what you use.
Sixteen AI families, each with its current models. Pick any combination per prompt, from one to all sixteen.
Logos are trademarks of their respective owners, shown to identify the models available in byself. byself is not affiliated with or endorsed by these companies.
byself reads its model list live from OpenRouter. When a company releases a new model, it appears under that company’s tile as soon as OpenRouter lists it, for every account. There are no tiers that hold the newest model back.
At the providers, the newest model usually sits behind the top subscription of each one. In byself every account can use every model and pays only for what it actually uses.
Sending a prompt to every model, browsing history, the account and payment pages, and PrivacyPlus in action.
A native desktop app. Install it, sign in, and your first side-by-side takes under a minute.
Signed and notarized by Apple. Open the .dmg, drag byself to Applications, done.
Download .dmgA separate build for Intel-based Macs is planned after Windows.
Coming soonEarly-access build. It isn't code-signed yet, so Windows shows two warnings; the download window walks you through both.
Prefer to read first? Open the overview (PDF) · Want to be told when your platform is ready? hello@byself.ae
Ask once and read every model together, decide who sees your data, and keep your whole history on your own machine.
Your prompt lands in every selected model at the same time. Raw answers appear side by side — no blending, no summary in between. Step through them one at a time, or scan them all together.
A green or red dot on every model shows its data-retention status at a glance. Flip PrivacyPlus on and the data-retaining models drop out of your selection, so they never receive the prompt at all.
Every session is stored in a database on your own machine — searchable, star-able, and exportable to PDF. byself's own servers never hold your conversations.
Your admin connects your OpenRouter account once, and every member gets their own private key. Remove someone and their access ends straight away. Admins manage seats and plans in one place, your organisation owns the billing, and every prompt goes straight from each member’s machine to OpenRouter.
byself is built in the DIFC, Dubai — for people who need to know exactly where their prompts go.
Requests go only to a fixed allowlist of no-training providers — never to unlisted hosts.
Your conversations live in a database on your own computer — not in our cloud.
Every model's data-retention status is shown before you send, so you stay in control.
byself routes each request through a controlled path to a fixed set of no-training providers. Chinese-origin open models run only on Western inference hosts. The diagram shows every host a prompt can reach. The detail for your IT team is one click below it.
A prompt leaves your device and goes to OpenRouter, which forwards it to the one provider that serves the answer. byself’s own servers are not in that path, so byself cannot read, store or train on your prompts. That leaves two parties who do see them.
OpenRouter states in its privacy policy that it does not use inputs or outputs for model training, and by default keeps only metadata (token counts, latency), not the text. byself has not enabled prompt logging.
The providers. byself requests carry data_collection: deny, which tells OpenRouter to route only to providers it classifies as not training on prompts. OpenRouter’s default is allow; byself sets it on purpose. Every request carries this setting, Cohere included: Cohere models are served from Cohere’s own infrastructure, and OpenRouter routes to them only because Cohere’s declared policy for that traffic does not permit training. Cohere models that OpenRouter serves under a zero-retention policy show green; the rest show red, may keep a copy for a limited period, and PrivacyPlus removes them. The closed-model companies in the lineup publish a no-training commitment for API traffic. Open-weight models such as DeepSeek, Qwen and Kimi are run by the Western hosts on the diagram, which publish no-training or no-retention commitments (CoreWeave is covered by OpenRouter’s classification rather than its own page); the model companies themselves never receive the prompt.
Training is not the same as retention. Some providers keep a copy for up to 30 days for abuse monitoring, then delete it. The ZDR badge in byself shows which models are served by zero-retention endpoints, and PrivacyPlus lets you remove the rest.
openrouter.ai. byself’s backend (Firebase) is contacted for sign-in, credits and key handling, and never receives prompt text.data_collection: deny on every model, the provider allowlist in provider.only, and allow_fallbacks: true, which lets OpenRouter retry another host on that allowlist only, never one outside it.What this does not prove: what happens inside a provider’s data centre. byself relies on the providers’ published terms, linked and dated here, and on OpenRouter’s classification of them. Those are contractual commitments, not something byself can inspect.
Pick up any past conversation exactly where you left it.
Turn any session into a tidy PDF — handy for reviews and audits.
Full right-to-left support across the workspace, not an afterthought.
Teams connect their OpenRouter account once; each member gets their own key, and you keep full control of spend.
Individuals top up credits; spending never exceeds the balance — no accounts to juggle.
A real Mac and Windows app, built for long working sessions.
One way for individuals, one for teams.
Buy credits and start comparing. Spending can never exceed your balance. No subscription, no provider accounts to set up.
Or any amount from $20. Credit value; a service fee is shown at checkout before you pay.
Run on your own OpenRouter account, with a private key for every member. Manage seats in byself and set spending limits per key in OpenRouter. Subscribe inside the app, no sales call.
Two additions in progress, plus the remaining desktop builds. No dates promised; each ships when it is right.
For credit accounts: open byself on your phone or any browser, pick a few models and compare on the go. Chat history stays on each device by design, so desktop and browser histories remain separate. Credit accounts come first; team accounts can follow, because every member now has their own key.
Tap the microphone in the prompt bar, talk, and your words appear as text ready to send. It will use a transcription model chosen for accuracy in Arabic as well as English.
Windows is out as an early-access build; a code-signed version will replace it on the download page. An Intel Mac build is planned after that.
Download byself for Mac or Windows and run your first side-by-side in under a minute.