EVE AI Core
EVE AI Core
From one real workflow to signed, hash-chained decision evidence you can verify offline — in roughly 60 days. This is the mechanics of an engagement, stage by stage: what gets deployed, what runs when, and exactly which artifact you walk away with at each step. No customer logos, no slideware — proof of process.
A pilot is a scoped, time-boxed deployment on a single real workflow in your regulated
domain — the one where a decision has to be defensible after the fact. It is not a sandbox demo on
synthetic data. CoreGuard evaluates each proposed action against a policy pack, returns an
ALLOWED, BLOCKED, or MODIFIED disposition, and emits a signed
decision record for every call.
The enforcement is deterministic: the same input reproduces the same governed result, so the evidence is reproducible rather than asserted. The commercial terms — founding-partner rate, duration, and how the pilot fee credits against an annual license — live on the design‑partner page. This page is about how the work actually runs.
Timings are the typical rhythm of a roughly 60-day engagement, not a contractual schedule — the workflow sets the pace.
We pick the single workflow together, define the decision schema CoreGuard will see
(proposed_action, model_output, context), select the policy
pack for your domain — lending_v1, insurance_v1,
healthcare_v1, government_v1 — and agree what success looks like
before any code is wired in.
CoreGuard goes in as an EVE Sidecar or via the API/SDK, running in
logging_only mode. It evaluates every decision and records the disposition it
would have taken — but it does not block your traffic yet. From day one you are
accumulating real, signed decision records on your own data, with zero risk to production.
We review the shadow records together, tune the policy pack against decisions that actually
occurred, drive down false positives, and promote the rules you trust from advisory to
enforcing. The calibration report is computed on your decisions, not
on a vendor benchmark.
Enforcement goes live: BLOCKED and MODIFIED dispositions now take effect
inline, within the contractual governance-latency target (<10ms p95 on the Enforcement tier). You
then independently verify the evidence offline — recompute the SHA‑256
content hash and check the Ed25519 signature with the public key, no EVE server in the loop.
eve-coreguard, eve-governance, eve-proof — all on PyPI.The auditor/regulator path needs no CoreGuard client at all. Install the standalone verifier, recompute the content hash, and check the signature against the published public key:
The CoreGuard SDK exposes the same primitives directly —
verify_decision_record(), fetch_public_key_pem(), and
recompute_content_hash() — and EVE Proof certificates verify the same way via
verify_certificate(). The full mechanism is documented in the
Python SDK reference and the assurance page.
A pilot is complete when four things are true — and not before:
No vanity metrics, and no public reference is ever a condition of the program. You can run a full pilot and remain entirely private.
If you operate a regulated workflow where decision evidence has to exist at the moment of the decision, a scoped pilot is the fastest way to prove it — on your data, verified by you.