// Value

A highly structured
Graph is the backbone.

Every model, every inference, every byte stays inside the perimeter — a private graph of operators, not a shared cloud.

// Embedded intelligence

The working interface
for Embedded Intelligence.

The verticallm platform brings your data, documents, processes and operational context together in one secure environment. Its interface gives every user a clear, role-specific view of what matters: developments, exceptions, risks, explanations and recommended actions. Instead of searching across systems or interpreting generic AI output, teams work with an intelligence layer that understands their company and supports the decisions they make every day.

// 01 · workspace
Good morning, Bas.
67 saved threads · context loaded.
Ask anything…
+  📎  </> ollama-local
Favorites Show all ›
Daily update
Learn more about creating beautiful websites.
#
Resilience report
Explore frameworks to speed up your workflow.
Workspace update
Making websites accessible to a wider audience.
DAILY UPDATE · JUL 16
NL plant · 07:00
signals scanned  27
critical         2
shipments on time  96%
OTIF · 7 DAYS
Open report · PDF Send to team
Your day starts briefed

Click a briefing: it reasons step by step, then hands you the report — with the numbers, the chart, and the actions.

// 02 · map & flows
All layers Customer  11338
3 customers isolated
A · 4 pending
A · 2 pending
A · 6 pending
Ask anything…
ollama-local
CUSTOMER · A-SEGMENT
Van Dijk Foods
customer_2000107 · Utrecht
pending orders  4
product        00321.TR
open value    €48.2k
ORDERS
so_0004212 · 12 pal · Jul 18pending
so_0004218 · 6 pal · Jul 21pending
so_0004224 · 9 pal · Jul 24pending
One living map

Ask the map in plain language: it reasons, isolates the customers, and one click opens the record.

// 03 · ingestion
↥  Drop files or bulk upload
HCCAP guideline.pdfdone
SLA-customer contract.43112.pdfdone
ISO-9001-manual.pdfdone
company-process-guide.pdfdone
📄
company-process-guide.pdf
company-process-guide.pdf
metadata complete
METADATA TAGS
knowledge layerGOVERNANCE
topicsCOMPLIANCESAFETY_RISK
primary topicQUALITY
confidenceVERIFIED
pii presentfalse
GRAPH & CHUNKS
vector chunks58
entities (indexed)37
relations (indexed)30
pipeline statusprocessed
Drop documents, get intelligence

Drag a file in: chunking, validity checks and KPI linking run live — click the result for its full metadata record.

// 04 · knowledge graph
nodes 14,321 · edges 43,121
supplier_1000014
entity  NODE
class   LOCATION
chunks  665
docs    12
CORPUS HEALTH
93.0%
The backbone

The knowledge graph as the backbone: click any node for its record, watch corpus health climb.

// 05 · kpis
OTIFservice
last value94.2%
validated
fill_rateservice
last value96.1%
validated
supplier_lead_timesource
last value6.3 d
validated
forecast_accuracyplan
last value87.0%
loaded
production_yieldmake
last value92.4%
validated
gross_margincash
last value31.8%
loaded
KPI · kpi.service.001
OTIF (On Time In Full)
DEFINITION
(on_time AND in_full) / total shipments × 100
LAST VALUE
94.2%Jul 15, 2026 · 30d window
HISTORY · 7 RUNS
LINKED KPIS
perfect_order_rate fill_rate order_cycle_time
Definitions that calculate

The KPI library at a glance: click one for its definition, history, last value and linked KPIs.

// 06 · nl→sql
Ask anything…
SQL agent · 6 tables joined
SELECT c.name, COUNT(d.id) AS late
FROM “sales_order” s
JOIN “delivery_order” d ON d.so_id=s.id
JOIN “customer” c ON c.id=s.customer_id
JOIN “product” p ON p.id=s.product_id
JOIN “supplier” su ON su.id=s.supplier_id
JOIN “country” co ON co.id=su.country_id
WHERE p.code=’00321.TR’ AND d.late
GROUP BY c.name;
Syntax
Safety
Schema grounding
Query detail
chat_87e54bb9_3149253
QUESTION
Which A-customers received late deliveries of 00321.TR from Dutch suppliers?
VALIDATION
Syntax    Safety    Schema grounding
EXECUTION
Result: success · Rows: 12 · Time: 447ms · Runs: 1
RESULT SAMPLE
customerlate
Van Dijk Foods4
Bakker BV3
Promote to benchmark ⚑ Flag for review
Ask in language, answer in SQL

A six-table question in plain language: the SQL builds step by step, gets validated, and every query keeps its full test record.

// 07 · evaluation
RETRIEVAL
Recall@k.92
MRR@k.88
Context relevance.94
GENERATION
Faithfulness.96
Answer relevance.91
RAG performance report →
EVALUATION · LAST 18 RUNS
RAG performance report
OVERALL SCORE
.93▲ +2.1% this month
KPIS
Recall@k         .92
MRR@k            .88
Context relevance .94
Faithfulness     .96
Answer relevance  .91
BY LAYER
retrieval .91 · generation .94 · SQL .91
18 runs · 30d window · next run 02:00
Open full report · PDF
Quality, measured

Retrieval and generation quality calculated live — the report opens with overall score, KPIs and layer breakdown.

// 08 · traces
Ask anything…
Current OTIF is 94.2% against the 95% SLA floor[1]. Below 95%, a 2% service credit per affected shipment applies[2], capped at 8% of monthly invoice value[3].
SOURCE · CITATION [2]
Article 7.2 — Service Credits
document  BP.SLA-2023.pdf
contract  SLA-customer 43112
date      27-07-2023
page      14 · chunk 137
PASSAGE
“…should OTIF fall below 95%, Supplier shall credit 2% of invoice value per affected shipment…”
every answer · every source · traced
Watch it cite

Ask about performance and liability: the contract agent answers with citations — click one and the exact SLA passage opens, article, contract and page included.

01 · The interface

One surface for the whole operation

Body: Everything happens in one workspace: ask a question, read the morning briefing, watch signals come in, or work directly on the map of your network. The system shows its reasoning step by step, and every answer carries citations — one click opens the exact contract clause, table or document passage behind it. No black box, no second screen. What you see is what the system knows.

02 · The platform

A system that runs on your floor

Underneath sits Datagraph: a knowledge graph of your entities and flows, a governed document corpus, and specialised agents for retrieval, SQL and contracts. KPIs are formally defined, calculated and validated; retrieval and generation quality are measured continuously. It all runs on your own hardware with local models — swappable as they improve, without your data ever leaving the building.