Everything you need
to know about Thurin

Thurin is a private AI platform — your own AI models running on dedicated infrastructure, not shared with anyone else. Think of it as having a personal AI assistant that lives entirely within your own environment.

Unlike consumer tools like ChatGPT or Claude, Thurin does not log your conversations, does not use your data for training, and does not run on shared GPU pools where other users' sessions could be adjacent to yours.

The key difference is infrastructure isolation. Consumer AI tools run on multi-tenant systems where your data touches shared pipelines — and the companies operating them have broad rights to use that data.

Thurin runs on dedicated GPU instances allocated to you. There is no shared pool, no adjacent tenant risk, and no platform-side logging of your conversations.

• Consumer AI: shared infrastructure, data retained by the provider, training data collection possible
• Thurin: dedicated compute, no conversation logging, open-source models with transparent weights

Thurin runs exclusively on open-source models — primarily from the Qwen family (Alibaba) and the GPT-OSS family. All model weights are publicly auditable, meaning there are no hidden behaviors or undisclosed training practices.

This is a deliberate choice. Open-source models give you full transparency into what you're running — you don't have to trust us, because you can verify the model itself.

Every new account starts with $15 in free compute credit — no credit card required. After that, you pay for GPU time as you use it.

There is no monthly subscription required. You are billed only for the actual compute time consumed while your model is actively generating a response.

When you send a message, a GPU starts generating your response. You are billed for the duration of that active computation — measured in seconds, priced by the minute.

If a response takes 2 seconds to generate, you pay for 2 seconds of GPU time. If you sit idle between messages, nothing is charged. Pricing varies by model tier:

• Fast tier — lowest cost, ideal for quick back-and-forth
• Premium tier — mid-range, better quality for complex tasks
• Premium+ tier — highest quality, highest compute cost

Exact per-second rates are shown in your account dashboard.

No. Your free $15 compute credit is available immediately after account creation with just an email address. No credit card is required to access the platform during the free tier.

Once your free credit is exhausted, you'll need to add a payment method to continue. We'll notify you before that happens.

Not currently. Thurin operates on a pure usage-based model — you pay for what you use, when you use it. This is intentional: a subscription would mean paying even during months when your usage is light.

A subscription tier may be introduced later for users who want predictable billing, but it will always remain optional.

The idle timeout is approximately 1 hour. If no requests come in for 60 minutes, the GPU spins down and your model enters a cold state.

If you're in an active working session, you won't encounter cold starts — the timer resets with every request. Cold starts are most noticeable at the start of a new session after a long break.

In the current beta, scale-to-zero is the default behavior for all users. A "warm reserved instance" option — where your GPU stays allocated around the clock — is planned for higher-tier plans.

For now, the practical approach is to expect a brief wait at the start of each new session and plan accordingly. Once warm, the experience is instant.

Thurin offers three model tiers, each running on dedicated GPU infrastructure:

All tiers run open-source models. You can switch tiers per-conversation from within the app.

Yes. Every model available on Thurin is open-source with publicly auditable weights. The current lineup draws from the Qwen family and the GPT-OSS family, both of which have published model architectures and training details.

Open-source models are a requirement for Thurin — not an afterthought. If you can't audit the model, you can't fully trust it. We will not add closed-source models to the platform.

Yes. The open-source model landscape moves fast. We evaluate new releases on a rolling basis and add models that meet our quality and transparency standards.

Beta users will be notified when new models are available. If there's a specific model you'd like to see, email us at [email protected].

No. Thurin does not log conversation content at the platform level. Your messages and the model's responses flow through the inference pipeline and are not written to any Thurin-controlled storage.

Your in-app conversation history (if you want it) is stored client-side or within your own account — not in a central Thurin database that could be accessed or breached.

No. Thurin does not collect, store, or use your conversations for model training — ours or anyone else's. This is one of the foundational commitments of the platform.

The models running on Thurin are open-source and fixed — they were trained on public datasets before deployment. Your usage does not alter or retrain them.

No one at Thurin. Your conversations run on isolated GPU instances and are not captured by the platform. There is no admin dashboard, no support access, and no engineering backdoor into your session content.

This is a design constraint, not just a policy. The architecture does not route conversation data through any system where it could be accessed by Thurin staff.

No. Your input goes in, the model generates a response, and that response is returned to you. The conversation content does not touch any external systems, logging infrastructure, or analytics pipelines.

Standard account metadata (email address, usage totals for billing) is stored separately from conversation content and is never combined with it.

During the beta, you need two things:

• A beta access code — distributed in limited batches to waitlist members
• An email address to create your account

That's it. No credit card, no phone number, no identity verification. Your $15 free compute credit is applied automatically when your account is created.

Join the waitlist on the Thurin homepage. We release access codes in batches as infrastructure capacity expands. Waitlist members are prioritized in the order they signed up.

Beta access is intentionally limited — we're scaling carefully to make sure the experience is solid before opening it up more broadly.

The primary access method is the web app at app.thurin.ai — no installation required, works in any modern browser.

A CLI tool and an API endpoint are in development for users who prefer terminal-based access or programmatic integration. These will be available to beta users first.

Email [email protected] with a description of the issue. During the beta period, response times are best-effort — typically within one business day.

For bug reports, include the approximate time of the issue and which model tier you were using. Screenshots or error messages are helpful if you have them.

Not yet. During the early beta, direct email is the most reliable channel. We're evaluating community options (Discord, forum) as the user base grows.

Follow @UseThurin on X for platform updates and announcements.

Send it to [email protected] and we'll get back to you. Frequently asked questions get added to this page, so your question may help the next person.