How do we integrate EVE?

As an inline sidecar in the request path or via the decision API (POST /v1/decisions/evaluate). Your models and agent framework are unchanged; EVE evaluates the proposed action and returns a verdict.

Is there an LLM in the decision path?

No. The policy kernel is deterministic compiled logic. The same normalized input under the same policy version yields the same verdict every time — no temperature, no sampling.

How is a verdict verified later?

Every decision is an Ed25519-signed certificate over a canonical content hash; any party can verify it offline against the published public key, and replay the decision to confirm the same result.

A Deterministic Policy Gate in Your Request Path

A deterministic policy gate in your request path

Governance shouldn’t mean bolting a second probabilistic model onto your stack. EVE AI Core is a deterministic gate — sidecar or API — that evaluates each action against versioned policy packs with no LLM in the decision path, and returns ALLOW / BLOCK / MODIFY.

The Gap

Governance bolted on late breaks the request path

Adding a guardrail model to judge outputs adds latency, another probabilistic dependency, and a control that drifts run to run. Engineering teams want a governance layer that is deterministic, testable, and cheap to call inline — one that decides authority for an action rather than re-scoring text. That is a different architecture: compiled policy, evaluated before the action runs, fail-closed.

What You Get

Integrates with your existing models

Sidecar or API

Deploy as an inline sidecar or call the decision API; your models and agent framework stay exactly as they are.

Deterministic, no LLM

Policy is compiled to fixed logic; the same input under the same policy version returns the same verdict, with no model in the decision path.

ALLOW / BLOCK / MODIFY

A structured verdict against versioned policy packs — if modified, only the modified form goes forward.

Signed, offline-verifiable

Every decision emits an Ed25519-signed certificate; auditors verify against the published key with no EVE account.

How It Fits

One gate, versioned like code

Policy packs are versioned and the verdict binds to the version, so governance behaves like the rest of your codebase — reproducible, diffable, and testable. See the API reference, the sidecar deployment mode, and the platform overview.

Common Questions

FAQ

As an inline sidecar in the request path or via the decision API (POST /v1/decisions/evaluate). Your models and agent framework are unchanged; EVE evaluates the proposed action and returns a verdict.
No. The policy kernel is deterministic compiled logic. The same normalized input under the same policy version yields the same verdict every time — no temperature, no sampling.
Every decision is an Ed25519-signed certificate over a canonical content hash; any party can verify it offline against the published public key, and replay the decision to confirm the same result.

See it on one of your decisions

Bring one real decision — see it enforced and proven

We run one of your real decisions through the deterministic gate, show the ALLOW / MODIFY / BLOCK verdict, and let you replay it and verify the signature offline. Founding design-partner pilot from $27,500 (standard $37,500).

Book a Pilot Run the Live Demo

Related: Why EVE · Platform · EVE CoreGuard · EVE Proof.