Frequently Asked Questions

ChatGPT processes your documents on shared infrastructure, and your data may be used for training. ArcticAI spins up a completely isolated GPU instance for each chamber. Your documents never leave your dedicated environment, are never shared, and are never used for model training. When you destroy the chamber, the data is permanently deleted.

Your documents are uploaded directly to your dedicated GPU instance via end-to-end encryption. They are processed locally on that instance using open-source AI models. Our orchestration layer only handles metadata (chamber status, billing) and never has access to your actual document content.

When you destroy a chamber, the dedicated GPU instance is completely terminated. All data, including uploaded documents, AI model memory, chat history, and generated outputs, is permanently deleted. The underlying storage is wiped and the instance is decommissioned. This process is irreversible by design.

We use open-source models served via vLLM. Flash tier runs Qwen 2.5 7B, Turbo runs Qwen 3.5 35B (a Mixture-of-Experts model), and Ultra runs GLM-5 744B. All model weights are fully auditable and run entirely on your dedicated GPU instance — no data is sent to any third-party API.

ArcticAI's architecture is designed with HIPAA-friendly principles in mind: each chamber runs on isolated infrastructure with encrypted storage, no shared resources, and complete data destruction on termination. We are not yet formally HIPAA-certified — full HIPAA compliance also depends on your organization's overall security implementation, so check with your compliance team before handling regulated health data.

ArcticAI uses simple pay-as-you-go pricing. GPU time starts at $0.50/hr for Flash (7B model), $2.50/hr for Turbo (35B model), and $20/hr for Ultra (744B model). You only pay for the minutes your chamber is running. Add credit to your balance and start analyzing — no subscriptions or commitments required.

Flash tier chambers typically boot in about 5 minutes. Turbo takes around 8 minutes because it downloads a larger 25GB model at startup. Ultra takes approximately 15 minutes due to its massive 744B parameter model. We're continuously optimizing boot times through Docker image caching and host affinity — if you've used a host before, subsequent boots on the same host are significantly faster.

ArcticAI supports PDF, DOCX, TXT, and other common document formats. Files are processed on your GPU instance using OCR and text extraction. Maximum file size depends on your plan: 25 MB for Pay-As-You-Go, 50 MB for Starter, and 100 MB for Pro.

No. Our orchestration layer never touches your document content. We only manage metadata like chamber status, billing, and instance health checks. Your documents are uploaded directly to and processed entirely on your dedicated GPU instance. We have zero knowledge of what you upload or analyze.

Yes. When you have a document selected, type an instruction starting with 'edit', 'update', 'change', or similar keywords. The AI will modify your document according to your instructions and show you a diff of the changes. You can review and accept or reject the edits before saving.

We provision GPUs from a global network of data centers. Flash tier runs on single 24GB GPUs (typically RTX 4090 or similar). Turbo uses L40S 45GB GPUs for the larger model. Ultra requires 8x H200 141GB GPUs for the massive 744B parameter model. Instances are provisioned from data centers in North America, Europe, or Australia with high-bandwidth connections — see our Privacy Policy for details on data location.

Auto-destruct is a safety feature that automatically destroys your chamber after a set period (1, 4, 12, 24, or 48 hours). This ensures that even if you forget to manually destroy a chamber, your sensitive data won't persist indefinitely. Auto-destruct also stops billing, so you won't be charged for idle chambers.

Yes. Documents are encrypted in transit (TLS) during upload to your instance. On the instance itself, data is processed in-memory where possible. The instance's storage is ephemeral — when destroyed, there is no persistent disk to recover data from. We never store your documents on our own servers.

Closed-source models like GPT and Claude require sending your data to third-party servers (OpenAI, Anthropic). With open-source models, the AI runs entirely on your dedicated GPU — your data never leaves your instance. The models are also fully auditable, so you can verify exactly what's processing your documents.

Unused credit balance can be refunded within 30 days of purchase. GPU time that has already been consumed is non-refundable, as the compute resources are provisioned and billed in real-time. If you experience a technical issue that causes unexpected charges, contact support and we'll review your case.

Currently, chambers are single-user. Each chamber is tied to one account, and shared or multi-user access isn't supported yet. Reach out if team access is something you need — it helps us prioritize the roadmap.