Xlaude

Two paths to the same chat.

Big model inference is heavy. Instead of building a pipeline we rent and resell, Xlaude connects to compute you already have access to — on desktop, your GPU running Ollama; on mobile, a free API tier with your own key. Both routes bypass us entirely.

Desktop

Install Ollama once. Xlaude talks to it.

Ollama is the standard local AI runtime — it manages model weights, serves inference over a local HTTP API, and runs on Mac, Windows, and Linux. Install it from ollama.com (we link you there), pull a model, and Xlaude discovers it automatically.

Your prompt → Xlaude app → localhost:11434 → Your GPU → Reply back

Mobile

Your API key. Direct from your phone.

Mobile phones can't host large models, so you plug a free API key from OpenRouter, Groq, or Google AI Studio into Xlaude. Requests go straight from your browser to the provider over TLS. The key lives in your browser only. We never see it.

Your prompt → Xlaude app → Provider API → Reply back

A model family named for how far light carries.

Spark is a flame on your desk. Ember is a hearth. Pharos is the lighthouse at the far horizon. Each name maps to a class of open-weight model, from a 3B that fits a laptop to a 671B frontier mixture-of-experts.

Free · Local

Spark

Browser-local models running on your GPU via WebGPU, or any model you've pulled with Ollama. Underneath: Llama 3.2 1B/3B, Qwen 2.5 1.5B/3B, Phi 3.5 mini, Hermes 3 (unrestricted variant).

1B–8B paramsFast

Pro · Hosted via Ollama / BYO-key

Ember

The daily driver. Mid-size instruction-tuned models that punch far above their weight. Underneath: Llama 3.3 70B, Qwen 2.5 72B, Qwen 2.5 Coder 32B, or whichever next-gen 70B class is current.

32B–72B params128k context

Ultra · Frontier BYO-key

Pharos

Frontier-class open models — top of the stack. Reasoning, long-form writing, hard coding problems. Underneath: DeepSeek V3, Llama 3.1 405B, or current frontier open-weight.

405B+ paramsReasoning

What's actually in the app.

A chat tab you'd expect, a code tab you probably wouldn't, and a local agent that lets Xlaude actually touch your machine — with your consent, every time.

Chat Tab

Streaming responses, markdown, code blocks with copy, multi-model picker, system-prompt presets including Unrestricted, Writer, and Coder.

Code Tab

Open a folder, browse the tree, read files inline. AI sees the files you pick, proposes edits as diffs, you apply or reject. Nothing writes without your click.

iii

Local Agent (Ultra)

A Node binary that runs on 127.0.0.1 only. Shell commands, multi-file edits, git. Every command needs your approval unless you trust-the-session.

One payment. Lifetime access.

No subscriptions. No "lifetime deal expires in 48 hours." We don't run inference, so we don't have ongoing costs that need you to pay forever. Pay once, keep the app forever, including future updates to that tier.

Free

In your browser, no sign-up required.

Spark model family (WebGPU)
Unrestricted / Writer / Coder presets
Chat history in your browser
Export chats as markdown
Light & dark theme

Launch now →

Questions you'd be right to ask.

Does Xlaude read my chats?

No. On desktop, your prompts go from your browser to Ollama on your own machine — localhost traffic that never leaves your box. On mobile, requests go from your browser straight to the inference provider whose API key you supplied. We don't proxy either path. If chat sync is on, we store the chat history encrypted at rest — but we still cannot read the content.

Why open-weight models and not GPT or Claude?

Because the open-weight stack (Llama, Qwen, DeepSeek, Hermes) has caught up fast — DeepSeek V3 is top-5 globally — and because you can run them on your own hardware. We don't have the right to resell Anthropic or OpenAI responses, and we don't want you tied to anyone's weather.

What is Unrestricted mode and what are its limits?

Unrestricted mode strips the lobby-reflex refusals some open models were trained with. It is useful for fiction, offensive-security research, dark creative writing, and technical requests that trip overly-cautious filters. Three floor lines stay regardless of mode: no sexual content involving anyone under 18, no actionable targeting of specific real people or infrastructure for live attack, no actionable synthesis routes for mass-casualty weapons. Those are in the system prompt and in the Terms.

What does my $15 or $20 actually unlock?

Features of the app itself: the Code Tab, cross-device sync, Ollama connection UI, BYO-key support, and (Ultra) the Local Agent. Inference is still powered by your Ollama or your own API account — we're selling the frontend, not the compute. Which is why it's one-time pricing instead of a subscription.

Will my purchase ever expire?

No. One payment, access forever. If we ever launch a hosted-inference subscription on top, your one-time purchase is unaffected.

Can I refund?

Yes — 7 days, no questions asked. Email support@xlaude.xyz with "Refund" in the subject line. Full details in the Refund Policy.

Mobile — do I really need my own API key?

Yes. Phones can't host 70B+ models, and we're not spending our margin on hosted inference you'd then resell by sharing an account. The free tier at OpenRouter and Groq is generous enough that most mobile users never pay the provider either.

Is my data really not stored?

On the Free tier and in BYO-key mode: literally yes. Nothing is stored on our servers that you can't delete with one click. If you turn chat sync on (Pro/Ultra), encrypted history is stored to give you cross-device access; you can turn it off or wipe it anytime. See the Privacy Policy.

The AI that stays yours.