Product · Vera
Public beta · accepting pilots

One AI gateway.
Three deployments.

One Rust binary. Vera Edge on the laptop. Vera Hub on the network. Vera Cloud in your tenant. Sharing one policy, one audit chain, and one model registry across local open-weight models and frontier APIs. The request pipeline is mechanically verified in Lean 4.

Pipeline verified in Lean 4 Wasm-sandboxed connectors EU AI Act · Article 12
Live demo Apply as a design partner
The three deployments

Same binary. Same policy. Three failure domains.

Edge, Hub, and Cloud are not three products with shared marketing — they are the same compiled artifact, configured for three placements in your topology. A policy authored on a developer's laptop applies unchanged in the tenant VPC.

Vera Edge

On the laptop.

Fast. Fully private. No cloud dependency.

  • Ships with Moonshine (speech) and Gemma (LLM) for on-device inference.
  • SSNs and PII never leave the device — even when pasted into prompts.
  • Works fully offline. The flight-mode case is a first-class deployment, not a fallback.
Vera Hub

On the network.

Department-scale policy and orchestration.

  • Microsoft Presidio powers context-aware PII redaction inline with prompts.
  • Agents announce capabilities to the Hub; clients invoke them through the same pipeline.
  • Hosts the team's Ollama commons of open-weight models.
  • Proxies to public clouds only when policy allows it.
Vera Cloud

In your tenant.

The same gateway, scaled inside your VPC.

  • Multi-tenant deployment inside the customer's own cloud account.
  • One policy and one audit chain across Edge, Hub, and Cloud.
  • Proxies to OpenAI, Anthropic, and any OpenAI-compatible upstream.
Underneath

The load-bearing pieces.

Most AI gateways are a thin proxy with a UI on top. Vera's load-bearing pieces are the ones that determine whether the system is auditable, federable, and safe under load — and they are unusual because most AI infrastructure does not have them.

Verification
The request pipeline ordering — auth → policy → QoS → dispatch → audit — is mechanized in a Lean 4 proof. No code path can skip a stage; the type system enforces it.
Federation
One policy and one audit chain across local-llama, vllm-internal, openai, anthropic, and any OpenAI-compatible upstream. The application stack does not change when the model does.
Redaction
Microsoft Presidio runs inline with every prompt. Sensitive entities are replaced with stable placeholders, the model reasons on the redacted content, and the response is deterministically unmasked on the way back.
Sandboxing
Every model integration runs as a WebAssembly component with no direct network access. Vera mediates all egress through allow-lists and credential injection — connector code never sees a secret.
Audit
Every request, model selection, redaction step, and policy decision is emitted to a tamper-evident, hash-chained, append-only log. EU AI Act Article 12 traceability falls out of the design; it is not a feature added on top.
Agent registry
Vera Hub doubles as an agent registry. Agents announce their capabilities; clients discover and invoke them through the same verified pipeline as a model call.
How a request flows

Five stages. Every request. Every deployment.

The order is the proof. Any caller — CLI, web, agent, Veya — goes through the same five stages, in the same sequence, whether the request lands on Edge, Hub, or Cloud.

Authenticate

Arrives over TLS with mTLS, JWT, or API key. Vera binds the caller to an immutable Principal.

Policy

ACLs, quotas, content rules, and intelligent mask mode. Sensitive entities are redacted before they leave the device.

Route

Frontier model when policy permits. Local open-weight model when sovereignty requires. Same API surface either way.

Dispatch

The Wasm connector executes inside a sandbox. Egress only through Vera. Secrets never reach connector code.

Audit

Prompt and response are hash-chained into an append-only audit log. Article 12-aligned and tamper-evident.

Design Partner Program

Vera is looking for lighthouse partners.

Bring a concrete gateway use case — sovereign inference, regulated routing, audit-grade traceability — and shape the product around it. Free during the program. Direct partnership with the founder.

See the program Try the live demo