Origami AI
Intelligent Folds

The Infrastructure
Behind the Intelligence

Five pillars of encrypted infrastructure — compute, storage, databases, AI models, and an immutable ledger. Built from the ground up with sovereignty in mind.

Your infrastructure. Your keys. Your control.

Every Origami deployment runs on bare metal — not shared virtualization. Our custom KVM hypervisor isolates each customer at the hardware level. One customer per host. No exceptions.

Confidential VMs powered by AMD SEV encrypt data in use, not just at rest. Memory is encrypted. CPU state is encrypted. Even Origami operators cannot see what's running inside your enclave.

Orchestration is fully autonomous. Zero human touch between provisioning and operation. Infrastructure scales, heals, and optimizes itself — without anyone ever accessing your environment.

1 Customer Per Host
Complete hardware isolation. No shared tenancy.
AMD SEV
Confidential computing. Encrypted memory and CPU state.
0-Human Orchestration
Fully autonomous provisioning, scaling, and healing.
Custom KVM
Purpose-built hypervisor. Not a wrapper on existing tech.

Petabyte scale. Double encrypted. Four tiers of warmth.

Object and block storage built on MinIO and GlusterFS, erasure coded at 4-parity. Data is encrypted at the application layer before it hits disk, then encrypted again at rest. Two layers. Two keys. Two separate trust boundaries.

Four-tier warmth moves data automatically based on access patterns. NVMe for hot data. SSD for warm. HDD for cold. LTO tape for archive. Every tier is encrypted. Every transition is logged to the immutable ledger.

Scales to petabytes without architectural changes. No forklift upgrades. No re-architecture. Add capacity and the system rebalances itself.

4-Tier Warmth
NVMe, SSD, HDD, LTO. Automatic migration by access pattern.
Erasure Coding: EC 4
4-parity redundancy. Survives multiple simultaneous failures.
Double Encrypted
Application-layer encryption plus at-rest encryption.
Petabyte Scale
No forklift upgrades. Add capacity and the system rebalances.

Millions of writes per second. Postgres compatible. Your schema.

Our database is a fork of Citus — distributed Postgres — rebuilt for per-domain scaling. Each customer's data lives in its own logical domain with dedicated NVMe storage. No shared tables. No noisy neighbors. No multi-tenant data risk.

Dynamic scaling adds or removes nodes per domain based on load. Hundreds of nodes in a single cluster. Millions of atomic writes per second per domain. The system scales horizontally without application changes.

Full Postgres wire compatibility. Your existing tools, ORMs, and queries work without modification. The difference is underneath — distributed architecture, per-customer isolation, and encryption at every layer.

Citus Fork
Distributed Postgres, rebuilt for per-domain isolation.
100s of Nodes
Per cluster. Dynamic scaling based on domain load.
Per-Domain NVMe
Dedicated storage per customer. No shared tables.
Wire Compatible
Full Postgres compatibility. Existing tools work unchanged.

Analytical intelligence. Effectively unlimited context.

Origami's current generation models — Tani and Yama — are designed for data science and business operations. Not creative writing. Not general-purpose chat. Analytical intelligence that processes your data, finds patterns, and delivers structured insights.

Tani is a ~300 billion parameter Mixture of Experts model. Yama is a ~1 trillion parameter Mixture of Experts model. Both use ShiftLeft — our attention tiling mechanism that gives them effectively unlimited context windows. No retraining. No architectural changes. Applied post-training to any transformer.

Both models run on Origami's five-pillar infrastructure. Every inference is logged to the immutable ledger — inputs, outputs, and reasoning. Not just what the model said, but why.

See our full model history
Tani: ~300B
Mixture of Experts. Data science and business operations.
Yama: ~1T
Mixture of Experts. Our largest analytical model.
ShiftLeft
Effectively unlimited context. Applied post-training.
Fully Logged
Every inference recorded to the immutable ledger.

Every operation. Cryptographically recorded. Independently verifiable.

Built on SCITT — Supply Chain Integrity, Transparency, and Trust — the IETF standard for immutable, tamper-evident audit trails. RFC 9943. One of the first deployments globally. First outside Microsoft's ecosystem.

Every read, every write, every attempt — including failures — is signed using COSE_Sign1 and registered to a Transparency Service. The TS appends to a Merkle tree: append-only, linear, irrevocable. Each operation gets a receipt — cryptographic proof of inclusion that anyone can verify without contacting the service.

This isn't a compliance add-on. It's the architecture. HIPAA, SOC 2, GDPR audit trails are automatic. Every AI decision is logged with inputs, outputs, and reasoning. Two parties can never be shown two different versions of the truth.

SCITT (RFC 9943)
IETF standard for immutable, tamper-evident audit trails.
First Non-Microsoft
First SCITT deployment outside Microsoft's ecosystem.
COSE_Sign1
CBOR Object Signing. Every operation cryptographically signed.
Append-Only Merkle
Linear, irrevocable. Any alteration breaks the chain.

One infrastructure. Four brands.

Each brand serves a different audience. All run on Origami's five-pillar infrastructure. All benefit from the immutable ledger.

Looking for a turnkey solution?

Our trusted implementation experts handle everything — zero internal lift. Our partners deploy, customize, and support Origami products for enterprise customers.

Explore integration partners

Ready to build on Origami?

Let's talk infrastructure. Our team will get back to you within one business day.