OrchestrateThe harness

Processes that run themselves

Turn the work your team repeats every day into flows that read your data, decide, ask a person when it matters and act across your systems. People stop copying between systems, and every run is on record.

Supplier invoices· FlowOn

New invoiceWebhook

Read the linesAsk a model

rows

Match the orderQuery an explore

Over R 50,000?If / else

true

Finance approvesApproval · 24 h

Post to ledgerCall an API

Tell the buyerWhatsApp template

Schedules and starts

Webhook, with its own secret+ Schedule · + Webhook

Budget

AI steps spend Finance's budget

38% used this month

RunStep

Flows

Automate a process end to end, by drawing it

Pick from 33 kinds of step, join them on a canvas and name each branch. Query your data as the flow's owner may see it, branch and loop, write rows, load the warehouse, and call any system with an API. Step through a run to check each step's output before the next.

Start by hand, on a schedule in your time zone, or from a webhook
When a run fails, the assistant proposes a fix for you to look over
History of every version, and copy a flow to another workspace

Add a step33 kinds

Data

Query an exploreFilterAggregateMergeLoad into the WarehouseChange a row

Compute

Ask a modelAsk about dataAgentCode

Control flow

DecisionIf / elseFor eachWaitApproval

Outbound

Send an emailEmail a dashboardCall an APISend a textPhone callWhatsApp template

Approvals

People decide, and the work carries on

A flow can stop and ask a named person to approve, with a deadline. They decide in OQLIS or from a link in their email, and the flow carries on the moment they do. Automation handles the volume; people keep the judgement calls.

Named approvers and a timeout on every approval step
Approved and rejected each lead down their own branch
Viewers can be allowed to approve without building anything

ApprovalsFlows and agents waiting for your yes or no

Pay Kalahari Freight, R 84,300Waiting for you

From Supplier invoices · asked 09:12 · closes tomorrow 09:12

Invoice inv-4471 matches purchase order 2210 on all 14 lines. It is over the R 50,000 limit, so it needs Finance.

A note (optional)Checked against the delivery notes

RejectApprove

AI steps

AI where it helps, rules everywhere else

Use a model to read an invoice, sort an email or pick a branch, and keep everything else as plain, predictable steps. Every AI step goes through your gateway with your keys, and each flow spends a budget you choose.

Ask a model, ask about data, a decision between branches, or a full agent
Choose whose budget a flow's AI steps use
Run as a named service account, with that account's access

Decision· A model picks one of the branches

QuestionIs this email a complaint, an order or a question?

1

complaintOpen a case and alert the account manager

12 today

2

orderMatch the customer, draft the order

41 today

3

questionAnswer from the knowledge base, for review

27 today

BudgetCustomer service, R 1,500 a month22% used

Pulses

Agents that keep watch for you

Give a pulse a goal, the data it may use and what starts it. It checks on its rhythm, remembers what it found last time, and tells you only when something needs attention. It asks before it changes anything.

Start from "Warn before stock runs low" or "Catch orders that stall"
Each run's answer, tool calls and tokens are kept
Works as the person it runs as, with their row access

PulsesNew pulse

NameLatest findingStatusNext run

Stock running lowRundu and Katima below 7 daysNeeds attention10:00

Orders that stallNothing new since MondayWatching11:30

Unusual cost spikesTyres up 34% at Walvis BayNeeds attentionTomorrow

Weekly revenue dipsRevenue within 2% of last weekWatchingMon 07:00

Stock running low · I found two depots under 7 days of stock at this month's use. I've drafted transfer requests from Central; they wait for your approval.

Deliveries and alerts

The right report in the right inbox, every Monday

Email a dashboard every morning, every Monday or at month end, as a PDF, a picture, a deck, a spreadsheet or a link, drawn as each person who receives it may see it. Alerts check a number on a schedule and tell people by email, text, call or WhatsApp.

Send to people, to a role, or to outside addresses
Each recipient gets only their own rows
Alerts stay quiet for a while after they fire

Deliveries· Dashboards by email

Regional sales · Mondays 07:00Role: Regional managers · PDF · each sees their own region

Sent today

Board KPIs · month endBoard members · PowerPoint

30 Nov

Alert · Stock under 7 daysWatching

How oftenEvery hour, 06:00 to 18:00Explore: Stock by site

ThenText the depot manager, email OperationsQuiet for 1 day

Playbooks

Automate the websites that have no API

Sign in to a supplier portal, find the month's statement and download it: record it once in the assistant's browser, and pulses replay it exactly. A playbook never guesses. It stops at the first page that is not as recorded.

Recorded once in a conversation, published as a version
Limited to the sites you list
Try it before anyone can use it

Monthly fuel statementVersion 3

1

Open portal.fuelco.exampleSites it may reach: 1

2

Sign inAs recorded

3

Open StatementsThe statements page

4

Choose the monthInput: month

5

Download the PDFInto Docs, Fuel statements

Try itStops at the first page that isn't as recorded

Unify

Your own code, without the risk

For the integration nobody else supports, write Python or JavaScript. It runs in an isolated sandbox, reaches only the hosts you allow, and takes its keys from sealed secrets. Scripts stay plain files you can run anywhere.

A live view of CPU and memory, with a Stop button
Secrets are written once and never shown again
Run on a schedule, or as a step in any flow

Weighbridge syncCodeRunsSettings

import json, os, requests

url = "https://weigh.internal/api/readings"
key = os.environ["WEIGH_KEY"]

rows = requests.get(url, headers={"key": key}).json()
print(json.dumps({"rows": rows}))

Networkweigh.internal:443 Secrets{{secrets.WEIGH_KEY}} Last run

Returned 1,284 rows · 3.2 sEvery 15 minutes

Activity

Know every process ran, and why one didn't

One page shows how every flow, schedule and script has been running: how many runs succeeded, which steps fail most and how long each takes. It shows statuses, times and codes, never what a run read.

The last 7, 30 or 90 days
Emails that failed to send, listed on their own
Open any run and step through what happened

Activity7 days30 days90 days

Runs4,812

Succeeded98.6%

Failed67

Runs per day

FlowStep that fails mostFailuresLast

Supplier invoicesPost to ledger21Tue

Weighbridge syncCode9Mon

Also in Orchestrate

The plumbing that keeps automation safe and tidy.

WebhooksWays in for other systems, each signed with its own secret.
Shared schedulesOne timetable, such as month end, used by many flows.
Secrets kept apartKeys flows and scripts use, sealed and never shown again.
If a run failsDecide on each flow what happens when a run fails.
TrashDeleted flows can be brought back.
From chatTasks in Create can start a flow through its signed hook.

"OQLIS has helped us automatically consolidate data from multiple tool sets."
Selvan Govender, Head of Cyber Security, GijimaCustomer stories →

Next · Control
AI you can account for→ Discover
The data every flow reads→