Obsidian

Private AI Infrastructure

Your data.Forged in black.

Dedicated GPU instances running open-source AI. Your documents never leave your machine. Self-destructs when you are done.

0
Data shared
100%
Local processing
E2E
Encrypted
Live Analysis
vault://local

document

merger-agreement-draft-v3.pdf

142 pages · 3.2 MB · processed in 4.2s

Classification

CONFIDENTIAL

Entities detected

47 PII references redacted

Risk assessment

HIGH — contractual exposure identified

Compliance

HIPAA / SOX compliant

Processing

100% local — zero data egress

GPU isolated · zero egress · AES-256
Scroll

Architecture

Forged for privacy

Six pillars of isolation. Each one non-negotiable.

01

Dedicated GPU

Your own isolated GPU instance. No shared resources, no noisy neighbors.

02

Self-Destructing Vaults

Workspace and all data obliterated on completion. Cryptographic proof of deletion.

03

Open Source Models

Fully auditable LLMs. No black boxes.

04

Full Audit Trail

Every query and response logged immutably.

05

Zero Knowledge

We never see your documents. Ever.

06

Network Isolation

No egress. Your data stays put.

Process

The forging

Four stages. From raw silicon to obsidian glass.

MINE
I

Provision

Click 'New Vault'. Dedicated GPU spins up in ~60 seconds.

Raw material extracted from the earth. Your isolated instance, carved from silicon.

Stage 1 of 4
SMELT
II

Upload

Drag and drop documents. Processed locally on your instance.

Ore meets fire. Documents enter the crucible, never to leave.

Stage 2 of 4
SMELT
II

Upload

Drag and drop documents. Processed locally on your instance.

Ore meets fire. Documents enter the crucible, never to leave.

Stage 2 of 4
FORGE
III

Analyze

Chat with documents. Extract data. Generate reports.

White-hot intelligence shaped under pressure. Every insight hammered into form.

Stage 3 of 4
SHATTER
IV

Destroy

When done, destroy the vault. All data permanently deleted.

The glass cools and breaks. Nothing remains. Cryptographic proof of obliteration.

Stage 4 of 4
SHATTER
IV

Destroy

When done, destroy the vault. All data permanently deleted.

The glass cools and breaks. Nothing remains. Cryptographic proof of obliteration.

Stage 4 of 4

Specialist on demand

When AI isn't enough.

Your data stays in your vault — until you explicitly send a brief to a vetted specialist. One click, one bill, no marketplace to browse.

Legal

Counsel review on the clause the AI flagged.

Financial

A CPA's eyes on a number the model wasn't sure of.

Medical

A clinical reviewer for the chart you can't sign off.

Pricing

Pay as you go

No subscriptions. No minimums. GPU time billed by the minute.

FLASH

Qwen 2.5 7B

Fast extraction and quick analysis. Ideal for straightforward document review.

$0.50/hr
01

AIR

Qwen 2.5 32B

Deep reasoning across complex documents. Nuanced analysis and multi-step queries.

$3.00/hr
02

PRO

Qwen 2.5 72B

Full-precision reasoning. Expert-level document intelligence for demanding workloads.

$4.00/hr
03

MAX

Llama 3.1 405B

The most powerful open-source model available. Uncompromising analytical depth.

$10.00/hr

Maximum power

04

Start your first vault

Add credit and begin. No contracts, no commitments.

BeginNo credit card required

FAQ

Questions

Everything about Obsidian's security and privacy architecture.

ChatGPT processes your documents on shared infrastructure, and your data may be used for training. Obsidian 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 run the Qwen 2.5 family of open-source models (7B, 32B, and 72B parameters) and Llama 3.1 405B, served via vLLM, the fastest open-source inference engine. All model weights are fully auditable and run entirely on your dedicated GPU instance — no data is sent to any third-party API. You choose the model tier that fits your workload, from fast extraction to full-precision reasoning.

Obsidian'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) on request. However, full HIPAA compliance depends on your organization's overall security implementation.

Obsidian uses simple pay-as-you-go pricing with no subscriptions or minimums. GPU time starts at $0.50/hr for the FLASH tier (7B model) and goes up to $10.00/hr for the MAX tier (405B model). You only pay for the minutes you use. Add credit to your balance and start analyzing — no commitments.

By default, your data never leaves your vault. When you request a specialist, you explicitly select the artifacts you want to share, optionally apply auto-redaction over the bundle, and confirm the export with a typed phrase. Every artifact that leaves is recorded in your audit log, and the specialist accesses it through a signed link with a 30-day expiry that we revoke the moment the deliverable arrives.

Still have questions?

support@obsidian.expert