Governance execution & authority control plane

TenantSage decides what the AI is allowed to know.

It sits between identity, policy, and retrieval — so no ungoverned evidence ever reaches AI retrieval, generation, disclosure, or execution. Authority before retrieval. Evidence before generation. Approval before execution. Proof before release.

governance_trace.log
Identity verified
S0 · via customer IAM / OIDC
OK
Authority resolved
S1 · ReBAC join-chain, <1ms
OK
Policy evaluated
S2 · classification · retention · hold
ALLOW
Evidence boundary sealed
S3 · eebHash bound to policy state
SEALED
Governed retrieval & context
S4–S5 · strictly inside sealed EEB
LOGGED
Released & recorded
S6–S7, S11–S12 · ledger entry committed
DONE
User / Agent
→ Customer Identity & IAM
TenantSage
→ RAG / Database / Tools
→ LLM
→ Disclosure or action

TenantSage ensures that no ungoverned evidence reaches AI retrieval, generation, disclosure, or execution.



What customers buy

Five components, one enforced sequence.

Each request passes through all five, in order. Skip a step and the request is denied — not answered on a best-effort basis. Full detail on the architecture page.

01 · GATEWAY

Governance Gateway

Entry point for every request. Nothing proceeds without verified identity, scope, action, and purpose.

02 · DAR

Authority Resolution

Resolves verified identity into a hashed, replayable authority snapshot — deterministic, not cached.

03 · POLICY

Governance Policy Engine

Classification, purpose, retention, legal hold — evaluated before a single document is touched.

04 · EEB

Sealed Evidence Boundary

Retrieval is mathematically constrained to a boundary computed before search runs.

05 · LEDGER

Evidence Ledger

Every step writes durable, hashed proof. No durable proof, no allow decision.

+

Approval & Execution

Mutating actions pass through an additional approval-bound block before anything runs.


Not a replacement

What everything else does, and what TenantSage adds on top.

CategoryWhat it does todayWhat TenantSage adds
IAM (Entra, Okta)Verifies identity and permissions per systemComputes what evidence AI may use for this request
AI guardrailsInspects prompts and responses after the factStops unauthorized data from entering context at all
Vector databaseFilters by namespace or metadata conventionDerives that filter from a proven authority decision
RAG platformRetrieves documents, assembles contextEnforces that retrieval can't expand past the seal
GRC / audit toolsReviews policy and events after the factEnforces write-ahead — no evidence, no permission

Said plainly

What TenantSage doesn't do — and isn't, yet.

Guarantee the LLM won't hallucinate Replace IAM or database authorization Replace DLP, SIEM, or a data catalog Stop forged runtime credentials
Current status: not production-ready
  • Hard isolation between customers hasn't been proven with populated cross-customer tests.
  • Cryptographic sealing and key custody for the evidence boundary aren't finished.
  • Canonical, lifecycle, and concurrency evidence coverage is incomplete.

Accurate description, today: a governance execution platform with real architecture and a working candidate implementation — still proving its security invariants before production use.

The bottom line

Not another AI. The proof that the AI was allowed to know what it knew.

TenantSage sells authority and evidence — not intelligence.