EVE Trusted Services

The composed governance platform: deterministic decisions, cross-call agent controls, MCP binding, artifact scanning, shadow evaluation, and signed evidence — as one enforcement and proof layer.

Who this is for

Organizations that need more than a single guardrail: a governance and enforcement layer that decides before actions run, reasons across an agent's calls, binds approved-to-executed for MCP tools, gates file ingestion by content, lets policy owners test candidate policies safely, and produces evidence any party can verify.

The problem

Point tools leave gaps. A per-call check does not catch harmful sequences; a monitoring dashboard does not enforce; a policy editor without shadow testing is risky; logs without independent verification are not proof. You want these composed so a decision, its enforcement, and its evidence come from one coherent layer.

Supported capabilities

Trusted Services composes the EVE product surfaces, each grounded in the capability registry:

• CoreGuard — deterministic ALLOW / BLOCK / MODIFY before a governed action executes, with no LLM in the decision path.
• Agent Governance — signed monitoring envelopes, deterministic anomaly detection (z-score, EWMA, CUSUM), prohibited-sequence blocking, and deterministic budget controls.
• MCP Governance — cryptographic approved-versus-executed binding, server-derived authenticated identity, and distributed single-use authorization consumption (PostgreSQL-authoritative, validated in the documented test topology).
• Artifact Governance — content-based media detection, bounded safe parsing, deterministic scanning, and artifact-digest-to-decision binding.
• Shadow Policy — authenticated, non-authoritative candidate-policy evaluation with signed divergence reports.
• Red Team — a customer-runnable adversarial suite with signed results and zero unauthorized side effects on blocked attacks.
• Proof — signed decision evidence and independent offline verification over jcs-1.

Readiness varies by surface: CoreGuard decisioning, agent controls, MCP governance, artifact scanning, shadow, and red-team are PILOT_READY; signed evidence, offline verification, and jcs-1 are SUPPORTED; framework adapters other than the generic adapter are EXPERIMENTAL.

How it works

1. Actions and tool calls are routed through a governance composition root.
2. CoreGuard renders a deterministic verdict; agent controls, MCP binding, and artifact scanning apply where relevant.
3. Every governed decision can produce signed evidence bound to that decision.
4. Policy owners test candidate policies in shadow without touching the authoritative verdict.
5. Any surface's evidence can be verified independently and offline.

Technical example (Python)

from eve_coreguard import CoreGuardClient

client = CoreGuardClient(api_key="eve_sk_...")

result = client.evaluate(
    request_id="req-001",
    tenant_id="org_abc",
    proposed_action={"type": "mcp.write_record", "table": "ledger", "row": {"amount": 1000}},
    model_output={"decision": "write", "confidence": 0.9},
    context={"principal_id": "agent_3", "session_id": "run_9"},
    policy_set="banking_aml_v1",
)
print(result.decision.status, result.risk.score, result.risk.level)

Signed evidence example

{
  "decision": { "action": "BLOCK", "reason_codes": ["…"], "decision_id": "…" },
  "content_hash": "…",
  "canon": "jcs-1",
  "signature": "kms-ecdsa-p384-…"
}

Verifier example

import { verifyDecisionCertificate } from "eve-ai-governance";
const result = await verifyDecisionCertificate(envelope);  // offline; public keys only, no network
console.log(result.valid ? "VERIFIED" : "FAILED");

Deployment options

• Embedded (in-service): in-process governance inside the EVE service.
• Hosted: clients govern via a reachable endpoint.
• Sidecar: a co-located governance process, and the recommended path for hard tool-call enforcement.
• MCP-gateway: for MCP binding and distributed consumption.
• Sovereign/offline: evidence and verification without a network dependency.

In every mode, a service error fails closed — no SDK silently falls back to permissive behavior.

Readiness

CoreGuard decisioning, agent controls, MCP governance, artifact scanning, shadow evaluation, and the red-team harness are PILOT_READY. Signed evidence, offline verification, and jcs-1 canonicalization are SUPPORTED. Framework adapters other than the generic adapter are EXPERIMENTAL. The EVE SDK core overall is a RELEASE CANDIDATE WITH EXCLUSIONS.

Limitations

• Each composed surface carries its own limits: the pip client governs via hosted mode; anomaly detection is advisory unless wired to enforcement; MCP is at-most-once and not remote exactly-once; distributed consumption is validated single-host, not multi-host; artifact OCR/vision is off by default; shadow is non-authoritative and its estimates are modeled; red-team is representative, not exhaustive; mappings are descriptive, not certification.
• jcs-1 is a constrained RFC 8785 profile, not full compliance.
• Only the generic framework adapter is validated; others are experimental and wrapper-enforced adapters are bypassable by a direct underlying-tool call. Hard enforcement requires a gateway or sidecar.
• The clients are published on public registries: pip install eve-coreguard (PyPI) and npm install eve-ai-governance (npm).

Next step

Request a pilot to compose the surfaces you need — CoreGuard plus any of Agent Governance, MCP, Artifact, Shadow, and Red Team — in the deployment mode that fits your environment.