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 vault. Your documents never leave your dedicated environment, are never shared, and are never used for model training. When you destroy the vault, 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 (vault status, billing) and never has access to your actual document content.

When you destroy a vault, 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 to support HIPAA compliance. Each vault runs on isolated infrastructure with encrypted storage, no shared resources, and complete data destruction on termination. We provide BAA (Business Associate Agreement) for Enterprise customers. However, full HIPAA compliance depends on your organization's overall security implementation.

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 vault is running. Add credit to your balance and start analyzing — no subscriptions or commitments required.

Flash tier vaults typically boot in about 3 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, 100 MB for Pro, and 500 MB for Enterprise.

No. Our orchestration layer never touches your document content. We only manage metadata like vault 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. All instances are in US, Canada, or EU data centers with high-bandwidth connections.

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

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, vaults are single-user. Each vault is tied to one account. Enterprise plans support team collaboration features where multiple users can share access to the same vault. Contact us for details on team and organizational accounts.