Home/ Products/ Bud SENTRY/ Product Brief
Bud SENTRY overview
Product Brief · Layer 06 · Security & Governance

Bud SENTRY

Secure Evaluation and Runtime Trust — the security, governance & compliance plane of the Bud stack: zero-trust model ingestion, CPU-native Sentinel guardrails at 0.70ms p50, 160+ policies, RBAC and identity brokering, and one immutable audit trace across the entire agent lifecycle. Built in, not bolted on.

Product reference v1.0 July 2026 ~13 min read
01At a glance

One governance plane. Every model, every call.

A model pulled from a third-party source isn't just weights — it can carry executable scripts, obfuscated binaries, or embedded malware. SENTRY enforces a zero-trust lifecycle for every model and every call: verified at ingestion, contained in a sandbox, continuously monitored in production — with guardrails, RBAC, and immutable audit on the same plane.

Guardrail p50 · CPU0.70ms
Attack success & false refusalboth <20%
Throughput · one Xeon node4,400+ req/s
Native context65,536 tokens
CPU-native serving benchmarks · methodology in §06
What it is
The governance plane of the platform — every model artifact passes a zero-trust supply chain before it runs: sandboxed fetch, multi-format scanning, exploit and trojaned-weight detection, gated registry
Every request passes Sentinel — CPU-native guardrails for jailbreaks, prompt injections, PII, and content safety — on the way in and again on the way out
Around the guardrails: 160+ policies, RBAC with your existing IdP, red-teaming and 140+ evaluation benchmarks, FinOps controls, and tamper-evident audit
EU AI Act traceability as one trace — not a forensic exercise across five tools
What it is not
A bolt-on guardrail API — it is embedded in every layer of the stack, enforcing policy on every request
The same thing as Sentinel — Bud Sentinel is the guardrail engine inside SENTRY, one component of the plane
A GPU service — enforcement is CPU-native with no egress, so the same plane runs air-gapped and out at the edge
02Where it fits

Layer 06 of the Bud stack.

SENTRY governs everything above and below it — every agent action, every inference call, every model artifact crosses this plane. Models, agents, and tokens all answer to the same control plane.

Governs above — every Studio and Agent action is admitted, policy-checked, rate-limited, and logged by SENTRY; agent access to MCP tools crosses the same plane.

Guards below — Sentinel classifies the input and output edge of every inference call AI Foundry serves, and every model entering via Model Foundry passes the zero-trust supply chain.

03Capabilities, in full

Verify. Guard. Enforce. Audit.

Six capability groups — the model supply chain, the runtime guardrails, and the governance machinery around them, expanded to the specifics an evaluator needs.

01Zero-trust model ingestionEvery model download triggers the security pipeline: sandboxed fetch, multi-format scanning, exploit detection, gated registry · scans .pickle, .safetensors, GGUF, ONNX and more for executable scripts, obfuscated binaries, and embedded malware · trojaned-weight detection — only verified artifacts enter the registry, and their verified state is the baseline production is monitored againstbefore it ever runs
02Sentinel — runtime guardrailsJailbreak detection, prompt-injection defence, and content moderation across toxicity, hate, harassment, self-harm, violence, illegal, and regulated-advice categories · 23 specialised models across 33 variants, CPU-native via Resource Aware Attention · PII redaction across 11 regions, prompt firewalling, and egress controls · 65,536-token inputs classified natively — a single call, not a chunking pipeline0.70ms p50 · CPU
03Compliance enforcement160+ policies with SOC 2, GDPR, and EU AI Act-aligned controls, enforced on every request · exportable compliance reports for auditors · EU AI Act traceability as one trace across the full pipeline, not five tools with no shared data model · honest posture — certified vs. in-progress documented on the Trust hubaudit-ready by default
04RBAC, identity & FinOps controlsGranular role-based permissions across modules, projects, and APIs — scope access down to a single model or tool · identity brokering via Okta, Entra ID, and Google Workspace — OIDC, OAuth 2.0, SAML 2.0 · rate limits, budgets, and FinOps policy applied at request admission — cost governance on the same plane as securityyour IdP · day one
05Evaluation, red-teaming & monitoringBud Evals native — 140+ benchmarks give immediate feedback on accuracy, safety, and compliance during agent development · 300+ ready-to-use probes, from regex checks to LLM-based evaluators, plus built-in red-teaming before anything ships · continuous cluster, system, and traffic monitoring — drift from verified state raises the alarm300+ probes
06Immutable audit — one traceTamper-evident logs for prompts, responses, model versions, and every access event · one trace spans the full agent lifecycle — SENTRY admits the request and closes it with the complete, attributable record · failure diagnosis in minutes, not hours of correlating timestamps across 8–12 separate logging systemsfull lifecycle

Bud Sentinel is a component within SENTRY, not a separate platform layer — trained on 4.5M labelled samples, the largest open guardrails dataset in existence, and 2.3× faster on a laptop than competing guardrails on a $15,000 A100 GPU. Served from one binary through the Bud Guardrail Gateway. The engine has its own deep-dive: the Bud Sentinel product brief.

04How it works

Architecture & components.

A zero-trust supply chain for model artifacts, an eleven-step governed request pipeline, and a layered guardrail engine built on Resource Aware Attention — wired so every request carries identity, policy, cost, and audit context automatically.

The zero-trust model supply chain — six gates

01FetchSandboxed download, contained
02Scan.pickle, .safetensors, GGUF, ONNX
03DetectExploits, trojaned weights, malware
04GateOnly verified artifacts enter the registry
05DeployVerified state recorded as baseline
06MonitorDrift from verified state → alarm

Zero-trust means the pipeline runs on every pull — a model from Hugging Face gets the same treatment as one from an internal registry.

The governed request pipeline — eleven steps, one trace

When an employee interacts with an agent in production, the request flows through the whole platform in a single governed pipeline. SENTRY and Sentinel own four of the eleven steps — admission, input guarding, output guarding, and the closing audit record.

StepComponentWhat happens
01Bud StudioAuthenticates the user; packages the request.
02Bud SENTRYthis layerValidates RBAC; checks 160+ policies; logs for audit; applies rate limits and FinOps.
03Bud Sentinel · inputthis layerClassifies for safety, toxicity, PII, compliance in 0.70ms on CPU. Blocks violations pre-inference.
04AI GatewayRoutes by complexity, cost, SLO — local SLM or frontier model; prompt caching cuts cost 23–40%.
05Bud AgentExecutes the workflow; decomposes tasks; manages state and memory; coordinates other agents.
06Bud MCP FoundryGoverned, auditable connections to CRM, ERP, databases, APIs. Every access logged and policy-checked.
07Bud Runtime + LayerZeroServes inference; abstracts hardware; self-healing.
08Bud Sentinel · outputthis layerClassifies the response for safety, PII, toxicity, compliance before it reaches the user.
09Bud ScalerSLO-aware compute allocation; FinOps attributes cost per department, use case, agent.
10Bud SENTRY · auditthis layerLogs the complete lifecycle in a single unified trace — every step attributable.
11ART · backgroundCaptures the interaction as a training signal; feeds continuous SLM improvement.

Resource Aware Attention — the mechanism behind Sentinel

Existing guard models take a GPU-designed transformer and compress it until it fits on a CPU — landing at 334–3,855ms per classification, capped at 512 tokens. RAA is designed from the deployment envelope inward, and three principles govern it.

The envelope is an input

The target cache hierarchy, precision tier, and latency SLO are declared before training — the mechanism is shaped by its resources, not retrofitted to them through post-hoc compression.

Bounded by envelope, not length

No quadratic phase change with sequence length: long inputs mean more of the same cheap work. That is why 65,536-token classification is native rather than a chunk-and-vote retrofit.

Heads compose over one pass

A single attention pass supports many task heads in parallel — safety, PII spans, intent, routing. A chain of five classifiers becomes one shared attention cost with cheap decisions on top.

The RAA layer is one of several. The fast layer handles the firehose at single-digit-millisecond latency; behind it sit deeper scans — higher-capacity review for tail risk, specialised detectors for adversarial patterns, and escalation paths for ambiguous cases. The speed of the fast layer is what makes the deeper layers affordable at production volume.

05Deployment & compatibility

Guardrails everywhere — including the edge.

CPU-native means no GPU dependency and no egress — the same enforcement plane runs in the cloud, on-prem, air-gapped, on a laptop, and out at the edge.

RequirementDetail
ComputeCPU-only — no GPU dependency, no egress. Enforcement runs on the same fleet the application already runs on, at the latency budget it already has.
Server envelopeIntel Xeon 6972P — 4,400+ req/s at 512 tokens, p99 under 12ms at 50-way concurrency.
Long-sequence envelopeIntel Xeon Platinum 8272CL — 65,536-token inputs at 560ms p50; transformer baselines cap at 512 tokens.
Laptop / on-deviceIntel Core Ultra 7 268V (fanless) — ~1,500 req/s at 128 tokens: guardrails inside a desktop app, browser extension, IDE, or embedded agent, with no server call at all.
Edge~25ms on edge CPUs — 96× faster than competing guardrails on the same hardware.
IdentityOkta, Entra ID, Google Workspace — OIDC, OAuth 2.0, SAML 2.0.
ComplianceSOC 2, GDPR, and EU AI Act-aligned controls with exportable reports; certified vs. in-progress posture documented on the Trust hub.

Deployment modes

On-premHybridCloudSovereign / air-gappedEdgeOn-device

The full supply-chain pipeline and Sentinel guardrails run in every mode — air-gapped deployments give up nothing, because nothing depends on a GPU or an external call. The same 160+ policies are enforced identically in every tier, so moving from cloud to VPC to air-gapped means zero re-integration.

Identity & SSO

OktaEntra IDGoogle WorkspaceOIDCOAuth 2.0SAML 2.0

Identity brokering with the IdP you already run, on day one — RBAC scopes flow from it across modules, projects, and APIs.

Threat & content coverage

JailbreakPrompt injectionToxicityHate · harassmentSelf-harmViolence · illegalRegulated advicePII · 11 regions

23 specialised models across 33 variants cover the input and output edges of every LLM call — plus model supply-chain threats before anything runs.

06Proof & methodology

Every headline number, with its basis.

Accuracy is reported on four public benchmark suites; competitors were measured via their published runtime pipelines on the bare machine, while Sentinel was measured end-to-end over its serving interface — its numbers include the network and RPC path a real request traverses.

0.70ms
Guardrail p50 latency
How it's measuredp50 classification latency at 10,000 concurrent connections on CPU (Sentinel serving benchmark). Per-request, end-to-end on named silicon: 5.67ms on EPYC 7V13, 5.99ms on Xeon 8272CL, 8.39ms on a laptop i7 — all at 512 tokens, including the RPC path.
<20%
ASR and FRR — only occupant
How it's measuredAggregate across JailBreakBench, PIGuard, WildJailbreak, and the Qualifire suite: 15.97% attack-success, 14.92% false-refusal — the only evaluated model with both under 20%. The two rivals with lower ASR refuse 82–89% of benign traffic; best single slice: WildJailbreak at 6.5% ASR / 11.42% FRR.
678×
Faster on the same silicon
How it's measured5.67ms versus 3,777–3,855ms for transformer guard baselines on the same EPYC 7V13 server CPU at 512 tokens. On a laptop CPU (8.39ms), Sentinel is faster than every baseline running on an A100 GPU (18–19ms) — 2.3× faster than $15,000 of dedicated hardware.
4,400+
Requests/sec, one node
How it's measuredSustained throughput on a single Xeon 6972P at 512 tokens, 50-way concurrency, p99 below 12ms. A fanless laptop sustains ~1,500 req/s at 128 tokens; a 65,536-token input classifies at 560ms p50 on a Xeon 8272CL — an input length the baselines do not operate at.
ClaimMetricBaseline / boundaryConditions
0.70ms p50classification latency10K concurrent connectionsCPU-native serving
5.67ms per requestend-to-end latencyvs. 3,777–3,855ms baselinesEPYC 7V13 · 512 tok · incl. RPC
8.39ms on a laptopend-to-end latencyvs. 18–19ms baselines on A100i7-11370H · 512 tok
ASR & FRR <20%15.97% / 14.92%Aggregate of 4 public suitesonly evaluated model under both
4,400+ req/ssustained throughputSingle Xeon 6972P node512 tok · conc 50 · p99 <12ms
65,536-token inputnative classificationBaselines capped at 512 tokXeon 8272CL · 560ms p50
~$0.10 per 1Mclassification costvs. ~$24 on GPU239× cheaper
84.56% balanced accuracyaccuracy#1 across 4 benchmarkssame suites as ASR/FRR
15–18× cost-performance$0.50/hr CPUvs. $2–3/hr GPU guardrailsingle node · Intel Xeon · sustained load
124M per dayclassificationsproduction platform scaleevery agent action guardrailed · CPU pricing
Intel Corporation
“By optimizing Sentinel for Intel® Xeon® processors, Bud enables enterprises to scale AI safety with high performance and significantly lower total cost of ownership, without the cost and complexity of GPUs.”
Sangeeta Roy — Director, Global Partner Business Leadership, Intel
07Why Bud Optional

The only deployable operating point.

The comparables are transformer guard models — Prompt Guard 2, Prompt-Guard-86M, ArchGuard, PIGuard, ProtectAI V2 — plus the 3–5 separate governance tools around them. The category forces a choice SENTRY doesn't.

Where the category stands

  • Transformer guards land at 334–3,855ms per classification on the server CPUs operators actually run
  • Input capped at 512 tokens — long context means chunking, voting, and re-aggregation pipelines
  • The low-ASR models get there by refusing 82–89% of benign traffic — unshippable operating points
  • Acceptable latency only with a dedicated GPU in front of every classifier
  • Guardrails, compliance, PII, audit, and access control split across 3–5 tools with no shared data model

Where only Bud SENTRY lands

  • Only evaluated model with ASR and FRR both under 20%
  • Single-digit ms on commodity CPUs — faster than the baselines on an A100
  • 65,536-token inputs natively, no chunking retrofit
  • Zero-trust model supply chain before anything runs
  • Guardrails, RBAC, policies, monitoring, and audit on one plane
  • One unified trace across the full agent lifecycle
  • No GPU, no egress — air-gapped and edge-deployable
  • Trained on 4.5M labelled samples — the largest open guardrails dataset
Deployable operating point CPU-native RAA 65,536-token native Zero-trust supply chain One unified trace No GPU, no egress
08Who it's for Optional

Where the plane earns its keep.

Six situations where governance being native — rather than bolted on across five tools — changes what an organisation can ship.

Security engineering

Third-party models, safely

A model pulled from a public hub gets sandboxed, scanned across formats, and checked for exploits and trojaned weights before it ever runs — and its verified state is monitored in production.

Compliance & risk

EU AI Act traceability, one trace

The Act requires traceability across the full pipeline. When that pipeline is one platform with one immutable trace, the audit is an export — not weeks of correlating twelve systems.

PII-sensitive workloads

HR & customer data with no gaps

PII detection and redaction across 11 regions run natively on every call, input and output — no gap between the guardrail tool and the logging tool for records to slip through.

Financial services

MNPI-safe multi-agent work

Multi-agent financial analysis where every action is policy-checked and logged — memory leakage between agents becomes a controlled, auditable surface instead of an open risk.

Government & sovereign

Air-gapped, no GPU, no egress

The full enforcement plane runs CPU-native with zero call-home — sovereign deployments go live in weeks on infrastructure that never touches an external API.

Edge & product teams

Guardrails inside the app

~1,500 req/s on a fanless laptop CPU means guardrails ship inside a desktop app, browser extension, IDE, or embedded agent — with no server round-trip at all.

09Go deeper & next steps

The full benchmarks, in depth.

This brief is the reference. For the complete benchmark tables, the Resource Aware Attention argument, and the per-suite accuracy numbers, read the whitepaper.

Get started with Bud

Put your data on it.

The fastest way to see what an integrated AI operating system does for your enterprise is a proof-of-concept on your infrastructure, with your data.

01 Identify a use case where complexity, cost, or governance is a known pain point.
02 Joint discovery — Bud maps your AI pain points to platform capabilities.
03 POC in days, on your hardware, with your data.