The data layer for AI agents

Every system you run. One graph your AI agents can trust.

Kelindan ties SAP or Accurate, WhatsApp Business, Google Sheets, your HR system and your warehouse into one graph that never goes stale. Each agent sees only what it is allowed to see, and every answer traces back to the row it came from.

Connect your first source See how it works

41 connectors · p95 sync 1.9 s · data stays in Indonesia

SAPSAP S/4HANAInvoice · CDC 2 s

ACCAccurateJournal · API 4 min

WAWhatsAppChat · webhook 1 s

GSGoogle SheetsVisit · every 5 min

HRTalentaEmployee · every 15 min

WMSJubelio WMSStock SKU · webhook 3 s

Collectionsinvoices, chats

Supportorders, stock

Stockread and write stock

CustomerPT Mitra Pangan Sentosa← SAP S/4HANA · KNA1 100482

InvoiceINV/2026/09/0388← SAP S/4HANA · VBRK 9000184417

InvoiceINV/2026/09/0412← SAP S/4HANA · VBRK 9000188213

ChatBu Rina, 8 Oct 14:02← WhatsApp · wamid 3EB0C8F2

VisitSales visit, 7 Oct← Google Sheets · Sales Bandung sheet, row 41

EmployeeDimas Saputra, sales← Talenta · Employee ID 0417

Stock SKUBRS-PRM-5KG← Jubelio WMS · Rack C-14, Cikarang warehouse

JournalJU-2026-10-118← Accurate · October general journal

123× denied

Live graph · mitra-pangan

48,211 nodes · 132,904 edges

Point at a node to see its lineageTap a node to see its lineage

01 · Sources

Data comes straight from where it lives.

Connectors read changes as they happen: the SAP and Accurate change logs, WhatsApp webhooks, Google Sheets revisions. No nightly CSV exports, no scripts that run at midnight.

Sources · mitra-pangan workspace

SystemLag · VolumeStatus

SAP S/4HANAERP · CDC2 s1.24M rows/dayLive

Accurate OnlineAccounting · API4 min18K rows/dayLive

WhatsApp BusinessMessaging · webhook1 s31K messages/dayLive

Google SheetsSpreadsheet · every 5 minsyncing 64%12 sheetsSyncing

TalentaHR · API15 min412 employeesLive

Jubelio WMSWarehouse · webhooktoken expired3 warehousesNeeds action

02 · Graph

Three different rows. One and the same customer.

Kelindan matches names, tax IDs, phone numbers and addresses across every source, then ties the records that mean the same thing into one node. Your agents never need to know SAP writes it in capitals.

SAP S/4HANA · KNA1PT MITRA PANGAN SENTOSA · NPWP 01.482.551.6-424.000

WhatsApp · contactMitra Pangan (Bu Rina) · +62 812 •••• 7731

Google Sheets · Sales BandungMitra Pangan Sentosa, Jl. Soekarno-Hatta 112

Customer · PT Mitra Pangan Sentosamatch 0.98 · 3 sources

nodes: 48,211
edges: 132,904
entities tied this week: 312

03 · Agents

An agent asks. The graph answers, lineage included.

Agents read the graph over MCP or the API. Their permissions are checked on every query, and every number in an answer points to the node it came from.

Collections agentasked over MCP

Which PT Mitra Pangan invoices are more than 30 days overdue, and who contacted them last?

Two invoices are past 30 days: INV/2026/09/0388 for Rp48,250,000, due 5 September [1], and INV/2026/09/0412 for Rp31,700,000, due 9 September [2]. The last contact was on WhatsApp with Bu Rina on 8 October at 14:02; she promised payment by 15 October [3].

[1]INV/2026/09/0388SAP S/4HANAVBRK 9000184417 · synced 09:41:07
[2]INV/2026/09/0412SAP S/4HANAVBRK 9000188213 · synced 09:41:07
[3]Chat with Bu RinaWhatsApp Business8 Oct 14:02 · synced in 1 s

credit_limit withheld: outside the Collections agent's permissions

Permissions per agent

Every agent holds its own key.

Permissions are set per agent, per entity, down to the column. Sensitive columns can be masked rather than refused, so an agent keeps working without ever seeing an ID number or a salary.

Agent	Invoices	Customers	Chats	Stock	Employees	Salaries
Collections agentMCP · 3 tools	Read	Masked	Read	None	None	Denied
WhatsApp CS agentAPI · 5 tools	Masked	Masked	Write	Read	None	Denied
Stock agentMCP · 4 tools	None	None	None	Write	None	Denied
HR agentAPI · 2 tools	None	None	None	None	Read	Masked

ReadWriteMaskedDenied

Checked on every query, not once per session. Every refusal is written to the audit log.

Data teams at 140 Indonesian companies tie their systems together with Kelindan

Mitra Pangan SentosaFMCG distribution

Pelita Farma DistribusiPharmaceuticals

Lintas Samudra LogistikLogistics

Gemilang RetailRetail, 38 stores

Baja Prima FabrikasiManufacturing

Sumber Tani AgroAgribusiness

Lumbung Dana SejahteraConsumer finance

Griya Medika GroupClinics, 21 branches

Measured in production, not in a demo

4 min

the stalest data an agent reads

← down from 24 hours when agents read daily exports

11 days

on average until the first agent is in production

← down from 5 months building the pipeline in house

data leaks across permissions

← 2.1 million agent queries, July to September 2026

Agents are tested before release, on your own company's data

Every change to a permission, a connector or a prompt runs the same test suite. An agent whose score drops does not ship.

Test suitecollections-v14 · 1,240 questions · 9 Oct 2026passed

Metric	With the graph	Baseline
Correct answers with the right source	96.8%	81.2%
Refuses questions outside its permissions	100%	62.5%
Numbers match the latest data	99.1%	74.0%
p95 answer time	1.9 s	6.4 s

Baseline: the same agent and model, reading a daily CSV export.

From the teams running it in production

“Our first AI pilot failed because the agent was reading yesterday's export. Now the graph updates within seconds, and the collections team finally trusts the numbers.”

Hendra WijayaCTO, Mitra Pangan Sentosa

“The auditor asked where a batch number came from. We clicked the node and showed them the warehouse row. Two minutes, done.”

apt. Sri WulandariHead of Quality Assurance, Pelita Farma Distribusi

“Four agents, four sets of permissions. Kelindan was the only one that could mask columns per agent without us rewriting the pipeline.”

Rizky RamadhanHead of Data, Lintas Samudra Logistik

A specification you can audit

Data residency: Data centres in Cikarang and Batam. No data ever leaves Indonesia.
Permissions: Per agent, per entity, per column. Checked on every query.
Lineage: Every answer links to its source row and the time it synced.
Audit: Every agent query is kept for 400 days and can stream to your SIEM.
Encryption: AES-256 at rest, TLS 1.3 in transit.
Models: Customer data is never used to train any model.

Tie in your first source today. Your first agent follows.

Connect one system in 15 minutes. Our engineers stay with you until your first agent passes its evaluation.

Book a technical session