The isolation primitive for multi-tenant AI

Isolation
you can prove.

Confidential AI serving for sovereign, regulated, and multi-tenant deployments. Isolation you can prove, not just promise, on commodity hardware with no foreign hardware root of trust. Because tenants can safely share one model, serving runs faster and costs less.

Sharing model state across tenants leaks. That is why safe multi-tenant sharing has meant an enclave per tenant, or not sharing at all. isovert makes shared serving provably safe, in software.

1000 tenants · one model · commodity hardware.

What it unlocks

Three things safe sharing makes possible

Share one model across tenants, safely.

MHA · GQA · MLA + state-space (SSM) + trained hybrids

Provable per-tenant isolation, validated across every major model architecture: the transformer KV designs in production today, and across the class boundary into the state-space and hybrid models the field is adopting.

Scope: representative models per architecture class, single hardware configuration.

Lower serving cost.

Cold-start reduced up to ~80×

Because tenants safely share one warm model, transformer serving cold-start drops sharply. Measured up to ~80× on production GPU serving (vLLM, A100), scaling with prefix length. This is a consequence of safe sharing rather than a standalone product.

Scope: transformer serving, single node, eager execution path. Validating at fleet scale with design partners.

See the serving-cost story

Catch prompt-injection attacks.

~91% detection · AUC 0.978 (AgentDojo)

Activation-based monitoring flags real third-party injection attempts for review, adding an advisory line of defense to multi-tenant agents.

Scope: advisory, human-in-the-loop, single model. It detects. It does not prevent, and it does not stop an adaptive attacker.

All of it is provable. Machine-checked proofs (Lean 4 + Verus), data authenticated at rest by construction (AES-GCM, bound to tenant, slot, and context), and 16 filed inventions behind it.

Published researchMachine-checked proofs (Lean + Verus)16 filed inventions

Positioning

Between policy and hardware

Isolating tenants has meant choosing between two imperfect options.

Software policy

Namespaces, RBAC, tenant-id filters. Cheap and portable, but enforcement-based. One missed check and tenants leak into each other, silently.

isovert

Tenant isolation as a structural property with a machine-checked proof behind it. Runs on commodity hardware and composes with the TEEs you already run. The assurance of a structural boundary without an enclave per tenant.

Hardware

Trusted execution environments. Strong host-level isolation, but a certified-hardware dependency, and in practice one enclave per tenant. A cost wall that doesn’t scale.

isovert doesn't replace the host boundary. It adds the per-tenant layer that policy can't reliably enforce and hardware has made expensive.

Architecture breadth

Validated across
every major architecture

Validation crosses the architecture-class boundary into hybrid and state-space models. Isolation that already works on what comes next.

01

Transformer attention

Validated across the major KV-cache designs in production today: MHA, MQA, and GQA.

02

State-space / SSM

Carried across the architecture-class boundary into state-space models, a class outside transformer attention entirely.

03

Hybrid models

Demonstrated on a trained hybrid that mixes attention and state-space blocks, the direction frontier models are heading.

Scope: tested on representative models per architecture class, on a single hardware configuration.

Properties

Structural
per-tenant isolation

Software-defined

No foreign hardware root of trust required. Isolation is enforced in software and composes with hardware where you have it.

Commodity hardware

Runs on the machines you already have. No special hardware required.

Provable on demand

Isolation you can audit at any moment during operation, not only at startup.

Key-drop delete

Drop the key that encrypts a tenant’s data and it becomes unrecoverable. Deletion you can demonstrate.

Authenticated at rest

Every stored vector is encrypted and authenticated with AES-GCM, bound to its tenant, slot, and context. A wrong-tenant or wrong-slot read fails the authentication tag and is rejected by construction, before any data is returned.

Composes with your stack

Sits inside the TEEs you run, and is designed to compose with a post-quantum (ML-KEM) key layer. It complements those boundaries rather than replacing them.

Attacker-tested

Evaluated under adversarial conditions across 1000 tenants sharing one model. Scope: correctness on a single hardware configuration.

The surface

Four calls

Key binding lives at the session layer — never an argument a model can supply. The isolation boundary sits below the application, where it can't be reasoned around.

isolation.py
01ctx = isovert.TrustContext(key=tenant_key)
02ctx.bind(data)
03ctx.run(query) # operates inside the isolated context
04ctx.delete() # key-drop delete: drop the key

The stack

One substrate.
Products on top.

NablaDB

in development

Confidential multi-tenant retrieval

Llamesh

early

Isolated multi-agent platform

isovertThe isolation primitive. The foundation for everything above.

From Cittela, a deep-tech research entity.

Proofs & verification

Don't trust. Verify.

Reviewed prior to publication by independent practitioners. Reproducible. Open.

Who it's for

01

Regulated multi-tenant AI platforms

Demonstrate tenant segregation to auditors and customers without one enclave per tenant.

02

Sovereign / national cloud

Provable isolation without depending on a foreign hardware root of trust.

03

Confidential-computing teams

The per-tenant layer of confidential computing, in software. Isolate tenants inside the trusted boundary you already run — composing with your TEEs and key management, and adding the tenant-level isolation that one-enclave-per-tenant makes expensive.

Provable isolation.