Architecture · Independent evidence assurance

Local evidence + external sources + org assurance

BeProof is an independent AI agent evidence layer, starting on macOS. It does not replace or perform the work of MDM, EDR, identity controls, or SIEM. Policy evaluation stays on the endpoint; the org plane aggregates signed, metadata-only summaries and source-attributed external evidence where adapters exist.

Two tiers

Tier 1 — Endpoint evidence plane

BeProofApp.app + CLI on each Mac. Collects AI-agent evidence, evaluates Policy Pack rules locally, maintains a hash-chain journal, and signs exports. Journal payloads are encrypted at rest (AES-256-GCM, Keychain-bound key); inventory tables are metadata-only but not encrypted in the current release.

  • Local scan — Declared, Granted, Observed evidence layers
  • Policy Pack evaluation on device (CoverageGap-first)
  • SQLite evidence store + hash-chain journal (never uploaded)
  • Ed25519 signed exports and fleet summaries
  • Review workflow — endpoint source of truth for evidence state

Tier 2 — Org governance plane

Cloud control plane + admin console. Tenant-scoped inventory, fleet health, review queue, and policy distribution — without server-side scanning.

  • Tenant-scoped device registry + enrollment tokens
  • Signed metadata-only fleet summary ingestion
  • Fleet metrics — heartbeat, stale audit rate (POL-M02)
  • Sanitized review queue + analyst assignments
  • Policy bundle publish (pull model; MDM precedence unchanged)

System diagram

How external controls fit

External products remain the source of their own control decisions. BeProof preserves those outcomes as evidence and exposes missing or stale integration coverage.

Outside BeProof

Source products act

Endpoint, identity, and vendor-native controls make their own allow, warn, deny, approval, or remediation decisions.

Contract available · adapters planned

BeProof attributes

Evidence keeps the originating product, source event ID, timestamp, freshness, confidence, and privacy class.

Available / expanding

Auditors verify

Signed manifests, coverage gaps, and transformation metadata make the resulting evidence reviewable without implying complete visibility.

Evidence trust boundary

BeProof enforces a normative split between local raw evidence and shared org evidence. Missing data surfaces as a CoverageGap — never a silent pass.

ClassLocationExamples
localRawEvidenceOn the MacFull SQLite database, complete audit report, journal bodies, raw credential locators
sharedOrgEvidenceSigned upload onlyFleet summary allowlist fields, export manifest metadata — no secrets, no raw DB
externalSourceEvidenceSource-attributed metadataProvider event ID, source health, policy decision, allow/warn/block outcome — planned per adapter

Key data flows

  1. Enrollment — fleet admin creates a one-time token; Mac runs fleet enroll; device credential registers with tenant.
  2. Managed audit — endpoint evaluates policy locally, signs FleetSummaryPayload, uploads to control plane; dashboard and review queue update.
  3. Policy publish — admin publishes bundle; endpoint pulls on sync. MDM system overlay path wins when present.
  4. Review sync — analyst bulk actions in cloud; endpoint pulls decisions. Full evidence paths stay on device.
  5. External evidence ingest (planned) — a provider-neutral adapter appends source-attributed inventory, incident, policy, or enforcement metadata. It cannot overwrite local evidence or turn a missing event into a pass.

Operating modes

Personal

Tier 1 only — local audit and signed export. No cloud upload required.

Managed local

Tier 1 + MDM overlay. Org policy context without fleet enroll.

Managed fleet

Tier 1 + Tier 2 — enrollment, heartbeat, admin console governance.

Integration points

BeProof integrates with existing stacks — it does not replace them.

SystemRoleStatus
MDM / UEMPKG, PPPC, org-policy overlay deliveryPilot
OpenAI / GitHub admin APIsCloud grants + account attributionPilot default
Endpoint / agent securitySource-attributed incidents, policy decisions, and enforcement outcomesContract + fixture; production adapters planned
Agent identity / credential controlAgent-to-identity lineage, credential policy outcomes, and source healthPlanned (R4–R5)
Firebase AuthAdmin console sign-in (pilot)Available
SIEM / webhookNormalized summary streamingPlanned (M4)
EDR / XDREnrichment metadata attachPlanned (M4)

What BeProof is not

  • MDM or device management replacement
  • EDR, XDR, DLP, identity enforcement, or runtime blocking
  • Central log store or SIEM replacement
  • Server-side policy evaluation engine
  • Auto-remediation for AI agent misconfigurations
  • A source of inferred blocked or running states without explicit evidence

Next steps

For a 30-day pilot walkthrough, see the design partner program. For MDM rollout, start with managed deployment runbooks.