Features
Forged for privacy
Obsidian provides dedicated GPU instances running open-source AI models for private document analysis. Your documents are smelted on a GPU dedicated to your vault. Destroy it when you are done.
Process
The forging
Four stages. From raw silicon to obsidian glass.
MINE
Provision
Click 'New Vault'. A GPU dedicated to your vault is carved from raw silicon, usually in a few minutes and occasionally up to 20. The typical spin-up time per tier is shown when you pick a model.
SMELT
Upload
Drag and drop your documents into the crucible. They are transmitted over TLS to Obsidian and relayed to the dedicated GPU instance provisioned for your vault, where they are indexed and held for the life of the vault. We don’t store them on our servers.
FORGE
Analyze
Chat with your documents using open-source AI models running on your instance. Extract data, generate reports, and mine deep insights, with no third-party AI service in the loop.
SHATTER
Destroy
When you are done, shatter the vault. The GPU instance and its disk, with your documents, are deleted. The glass breaks. Its saved chat history is deleted too.
Architecture
Six principles
Behind every vault. Each one forged in black.
Dedicated GPU Crucible
A GPU dedicated to your vault, rented from independent hosts on a third-party GPU marketplace. The GPU is yours alone; the host machine’s CPU, disk and network may be shared with other renters.
Destroy When Done
When you destroy a vault, its GPU instance and disk are deleted at the GPU provider, along with the documents on it, and the vault’s saved chat history is deleted from our database. Irreversible by design.
Open Source Models
Fully auditable LLMs with no black boxes. We run open-source models — Qwen 3, Qwen 3.5, and GLM-5 — served via vLLM, an open-source inference engine. The model weights are public, so anyone can inspect them.
Audit Trail
Vault, document and specialist events are logged with timestamps and can be exported as CSV for your records.
Dedicated Processing
Our orchestration layer handles metadata and routing — vault status, billing, provisioning. Your document content is smelted on the GPU dedicated to your vault. No third-party AI service ever sees it.
No Third-Party AI
No third-party AI service ever sees your documents. Answers come from open-source models running on your vault’s own GPU, and your documents are never used for training.
Use Cases
Built for sensitive industries
When the documents are too important to trust to a shared AI service.
Legal Review
Analyze contracts, briefs, and privileged communications on a GPU dedicated to your vault.
Financial Analysis
Process earnings reports and financial models without sending them to a third-party AI service.
Medical Records
Review patient records and clinical data on a GPU dedicated to your vault. Built for confidential work.
HR & Compliance
Handle employee records, investigations, and sensitive HR matters in vaults you destroy when the work is done.
Due Diligence
Process M&A data rooms with hundreds of documents without handing them to a third-party AI service.
Government
Analyze sensitive (non-classified) public-sector documents on a GPU dedicated to your vault.
Light the forge
Add credit and begin. No contracts, no commitments.