FEATURES · 2026

Everything PartSentinel ships.

Four themes, nine capability blocks, every feature on a single page. Print it, hand it to procurement, compare it to your RFP. We have nothing to hide and a lot to ship.

54 features · 9 capabilities · 4 themes
Audit a reference →See pricing
JUMP TO THEME
I
FOUNDATION
Foundation
II
SENTINEL INSIGHT
Sentinel insight
III
COMPLIANCE & TRUST
Compliance & trust
IV
DELIVERY & SERVICE
Delivery & service
PLAN INCLUSION
Single auditQuarterlyContinuous
WHICH COMMERCIAL PHASE?

Four themes. Three phases. No feature removed.

Features are organised by technical theme below. Here is how they unlock across the Reveal → Map → Control commercial journey.

REVEAL · v0.x

Diagnostic on 50–100 SKUs

Mode 1 (Blind) or Mode 2 (Full). Activates: Foundation (ingestion · pipeline) + Sentinel insight (Score, hallucinations, leaks).

  • I · Foundation
  • II · Sentinel insight
MAP · v1.x

Catalogue · competitors · markets extension

Adds: Compliance & trust (AI Act dossier, governance) + engine extension across categories and locales.

  • III · Compliance & trust
CONTROL · v2.x

Continuous monitoring · Negative Knowledge

Activates the full set: Delivery & service (recurring reporting, alerts, dedicated engineer) + boundary statements and controlled publication.

  • IV · Delivery & service
I
FOUNDATION

Foundation

Catalogue in, pipeline running.

2 blocks · 11 features
01

Catalogue import

Upload your catalogue as CSV or Excel. We deduplicate, we validate the basics, you start auditing.

CSV / Excel import

Upload your catalogue as CSV or XLSX. Automatic SKU deduplication, basic schema validation (required columns, encoding, empty rows). Ready in seconds.

Single auditQuarterlyContinuous
02

Audit pipeline

Multi-model probing with deterministic seeds, calibrated prompts, and rate-limited concurrency. Every audit reproducible to ±1 score point.

62 / 342 model audit panel

Frontier APIs, open-weights on EU GPUs, sovereign deployments — see /models for the full list.

Continuous
10-model standard panel

Top frontier and open-weights models — anonymised in public marketing, named under NDA.

QuarterlyContinuous
4-model diagnostic panel

Compact panel for one-time audits — frontier-only, calibrated to your vertical.

Single audit
Vertical-calibrated prompt templates

Per-vertical prompts: aftermarket, electrotechnical, aerospace, chemicals.

Single auditQuarterlyContinuous
Multi-language audit

Each model queried in English plus its native language family. Hallucinations differ across languages.

QuarterlyContinuous
Deterministic seeds

Fixed temperature, fixed seed, fixed prompt template version. Reproducible to ±1 score point.

Single auditQuarterlyContinuous
Rate-limit-respecting concurrency

Per-provider concurrency caps, per-region routing, automatic backoff.

Single auditQuarterlyContinuous
Audit on-demand

Trigger an audit run from the dashboard. Results in 5–72 hours depending on scope.

QuarterlyContinuous
Scheduled audits

Quarterly (default) or monthly cadence. Drift charts auto-generated.

QuarterlyContinuous
Stratified sampling

By vertical, age, OE coverage, revenue contribution. Configurable per audit.

QuarterlyContinuous
II
SENTINEL INSIGHT

Sentinel insight

Score, hallucinations, leaks — the observation layer.

3 blocks · 18 features
03

Sentinel Score

Single 0–100 metric, deterministic, calibrated per vertical, reported alongside the industry benchmark.

Sentinel Score (0–100)

Deterministic aggregate of five sub-metrics, weighted per vertical.

Single auditQuarterlyContinuous
Five sub-metrics

Identification, cross-reference, application, specs, procedural — each scored separately.

Single auditQuarterlyContinuous
Per-vertical calibration

Weights immutable per vertical_version. Aftermarket, electrotechnical, aerospace, chemicals.

Single auditQuarterlyContinuous
Industry benchmark

Anonymised cohort median + P10–P90 range. k-anonymity > 5 enforced.

QuarterlyContinuous
QoQ drift charts

Score evolution per audit, per vertical, per sub-metric.

QuarterlyContinuous
Competitive Exposure score

How much competitor cross-references gain authority over yours, per audit.

QuarterlyContinuous
04

Hallucination forensics

Per-reference, per-model, typed hallucination tracking. Find the failure mode, not just the failure rate.

Hallucination typology

Invention, confusion, extrapolation, obsolescence, supersession — five distinct failure modes, each tracked and counted separately.

Single auditQuarterlyContinuous
Per-model heatmap

Which models hallucinate most, on which categories, for which references.

Single auditQuarterlyContinuous
Root-cause hypothesis

Inferred hypothesis: scrape source, training-data window, model family bias.

QuarterlyContinuous
Severity tagging

High / Medium / Low severity tags on every finding, configurable per family.

QuarterlyContinuous
Submit & resolve workflow

Mark a hallucination corrected, attach evidence, re-audit on the next run.

QuarterlyContinuous
Per-language hallucination delta

Same reference, same model, different languages — we report the delta.

Continuous
05

Cross-reference leak detection

Find confidential cross-references that surfaced in model knowledge — with probable scrape source attribution.

Confidential cross-ref exposure register

Documented list of confidential cross-references found in model knowledge.

Single auditQuarterlyContinuous
Scrape-source attribution

Probable source: competitor catalogue, leaked PIM export, scraped PDF — with confidence interval.

QuarterlyContinuous
Internal SAP / engineering revision watchlist

Watchlist of internal codes that should never appear in model responses.

QuarterlyContinuous
Pre-launch reference watch

Detect when models recall pre-launch references — leak from upstream supplier or internal PDM.

Continuous
Investigate workflow

Per-leak drilldown, per-model audit log, raw response with the leaked code highlighted.

QuarterlyContinuous
Add-to-backlog batch action

Bulk-create remediation tickets via Linear / Jira / Asana webhook.

Continuous
III
COMPLIANCE & TRUST

Compliance & trust

AI Act, governance, EU-residency.

2 blocks · 11 features
06

AI Act watch

We watch your AI Act exposure — we don't file for you. Sentinel-mode advisory: we observe, we surface signals, your counsel decides. See /ai-act for the full positioning.

Catalogue provenance trail

Source, format, version, ingestion timestamp, content hash for every reference in scope. The observation log, not a legal deliverable.

Single auditQuarterlyContinuous
Per-model audit log (7-year retention)

Every prompt, every response, every model version, every timestamp. EU-resident, append-only storage — yours to consult, ours to retain.

Single auditQuarterlyContinuous
AI Act exposure dashboard

Live read-only view of the indicators that matter for AI Act discussions: provenance coverage, model log integrity, IP exposure trend.

QuarterlyContinuous
IP exposure brief (executive)

Quarterly memo with conservative and aggressive IP exposure valuations. Internal use, signal-only, no legal positioning.

QuarterlyContinuous
08

Compliance & governance

EU-resident, 7-year retention, k-anonymity on the public index. Built for the buyer who has to defend it.

EU-resident infrastructure

Hosting in Frankfurt and Paris. No data leaves EU at rest.

Single auditQuarterlyContinuous
AES-256 at rest, TLS 1.3 in transit

Per-customer encryption keys for catalogue blobs.

Single auditQuarterlyContinuous
Role-based access (RLS)

Database-layer row-level security. Every read scoped to the owning organisation.

Single auditQuarterlyContinuous
7-year audit log retention

Append-only storage. AI Act Article 53(1)(d) reproducibility.

Single auditQuarterlyContinuous
k-anonymity > 5 on public index

Sentinel Industry Index suppresses any cohort < 5 organisations.

Single auditQuarterlyContinuous
GDPR DPA

Standard DPA available on request. See /dpa for the summary.

Single auditQuarterlyContinuous
SOC 2 Type II

In progress, target Q3 2026. ISO 27001 target Q1 2027.

Continuous
IV
DELIVERY & SERVICE

Delivery & service

Reporting, exports, engineer-led onboarding.

2 blocks · 14 features
07

Reporting & exports

Every audit ships in four formats so every stakeholder gets the slice they need.

Executive PDF

12-page report. Sentinel Score, top hallucinated refs, top leaks, AI Act dossier ID.

Single auditQuarterlyContinuous
Raw Excel export

Every prompt, every model, every response. The full audit data, ready for your data team.

Single auditQuarterlyContinuous
CSV per-cell export

One row per (reference, model, prompt, response) — for downstream pipelines.

QuarterlyContinuous
Drilldown dashboard (30-day access)

Per-reference drilldown, per-model scorecard, per-leak investigation panel.

Single audit
Drilldown dashboard (full term)

Permanent dashboard access for the duration of your contract. Multi-user.

QuarterlyContinuous
Per-reference scorecard

PDF scorecard for any single reference, requestable on demand.

QuarterlyContinuous
Custom report templates

Branded report templates with your logo, colours, executive summary structure.

Continuous
09

Service & onboarding

Audits are technical work. We staff them with people who have seen 1000+ industrial catalogues.

1 catalogue onboarding included

Engineer-led onboarding for your first catalogue: format mapping, validation, calibration.

QuarterlyContinuous
Single-reference live probe

Drop one of your own SKUs into the live probe — 6 / 342 models queried, full Sentinel Score in 90 seconds.

Single auditQuarterlyContinuous
Dedicated audit engineer

Named engineer for Continuous customers. Quarterly review meetings.

Continuous
FR / EN / DE support

Support in three languages. Average response time < 4 business hours.

QuarterlyContinuous
Custom prompt templates

We build custom probe templates for your specific verticals or compliance regime.

Continuous
On-site audit review

Annual on-site audit findings review. Travel included for EU customers.

Continuous
99.5% SLA

Production SLA on the dashboard. Sandbox is best-effort.

QuarterlyContinuous
TRY IT · 1 REFERENCE · 90 SECONDS

See every feature in action.

Drop one of your own references into the live probe — 6 / 342 models, full Sentinel Score, exposure brief. No sales call required.

Audit a reference →Talk to a product specialist