Skip to content
teamsleek.ai

AI agents and workflows · Singapore

AI agents and workflows that ask before they act

Agents read your documents and records, then draft the quote, the purchase order, the bill or the reminder. A person approves it before anything changes, and every step is logged in your own system.

Book a scoping call See the agent library

Activity log, one morning

Agents prepare, people approve

  1. 07:01 · Purchase order agent

    2 items below their reorder points at Tuas. Drafted PO-1187 to supplier S-207: S$4,915.90 incl. GST.

  2. 08:01 · Collections agent

    Drafted 12 payment reminders. Left out 2 disputed invoices.

  3. 08:45 · Operations manager

    Cut the bolts to 250 boxes. Approved PO-1187.

  4. 09:15 · Credit controller

    Approved 11 reminders. Held 1 to call the customer first.

  5. 10:05 · AP invoice agent

    INV-55021 is S$30.00 over PO-1162, above your S$20 tolerance. Held for finance.

  6. 11:40 · Finance manager

    Price rise confirmed. Approved and posted S$1,166.30.

drafted by agents
14drafted by agents
approved by people
13approved by people
held for a call
1held for a call

Sample data. Harbourline Supplies Pte Ltd is not a real company.

Anatomy of an agent

Six steps, and a person in the middle

Every agent we build follows the same shape. It starts on a trigger, reads only what its job needs, drafts the work, and stops for a person. Only then is the change made, and every step is logged.

Worked example: purchase order agent

  • Agent prepares
  • Person decides
  • System
  1. 01

    System

    Trigger

    Something happens in your system or your inbox.

    M8 anchor bolts at Tuas drop below their reorder point of 200 boxes.

  2. 02

    Agent prepares

    Reads, with least access

    Only the records this job needs, read-only.

    Stock by location, open sales orders, supplier prices and lead times.

  3. 03

    Agent prepares

    Drafts

    Prepares the record and shows its working.

    Draft PO-1187 to supplier S-207 for S$4,915.90 including GST.

  4. 04

    Person decides

    Approval

    A person approves, edits or rejects it.

    The operations manager cuts the bolts to 250 boxes and approves.

  5. 05

    System

    Acts

    The change is made exactly as approved.

    PO-1187 is saved at S$4,240.10 and sent to the supplier.

  6. 06

    System

    Logs

    Who proposed it, who approved it, what changed.

    Stock figures used, the edit, the approver and the time.

Sample data. Harbourline Supplies Pte Ltd is not a real company.

What every agent comes with

  • One named job

    A plain description of the work it prepares, and nothing else.

  • Least access

    Reads only the records its job needs, and writes only drafts.

  • A control level

    Draft for approval by default. You can tighten or loosen it per action.

  • A named approver

    The role that approves its drafts, plus a second approver above your limits.

  • A log

    Every draft, edit and approval, kept in your own system.

  • Tests before go-live

    Agreed cases, run on a test copy, signed off by you.

  • An off switch

    Switch it off and the work goes back to your team's manual path.

  • And the code

    The agent is part of the system we build, so its source code is yours.

Control levels

You decide how far each agent goes

Every agent starts at Draft for approval. You can keep it there, tighten it to Suggest, or let it act within a S$ limit you set, one kind of action at a time. Each agent's page says which of its actions, if any, can do that.
  1. Level 1

    Suggest

    The agent recommends. A person does the work.

  2. Level 2 Default

    Draft for approval

    The agent prepares the record and waits. Nothing changes until a person approves. This is the default.

  3. Level 3

    Act within limits

    Opt in per action, under a S$ limit you set. Payments, deletions and messages to customers or suppliers always stop at a person.

Sample data. Harbourline Supplies Pte Ltd is not a real company.

Always stops at a person, at every level

Payments, deletions, and messages to customers or suppliers. No setting turns this off.

Agent library

One job per agent

Each agent belongs to a module and does one job. Filter by module or industry, or open an agent to see its trigger, access, limits, approval step, log and tests.

Module

Industry

Showing 6 of 6 agents

Workflows

Where agents prepare and people approve

Two everyday processes, step by step. Dark steps are prepared by an agent. Light steps are decisions a person makes. Dashed steps are your system doing what it already does.

Procure-to-pay

From stock running low to paying the supplier. Agents draft the PO and match the invoice; people approve the order, the bill and the payment.

  1. System

    Stock falls below its reorder point

    Checked every morning at a time you set

  2. Agent prepares

    Draft the purchase order

    Purchase order agent: supplier, quantity, price, lead time

  3. Person decides

    Approve and send the PO

    Buyer, or a director above your S$ limit

  4. Agent prepares

    Read and match the invoice

    Document intake and AP invoice agents, against the PO and goods received

  5. Person decides

    Approve and post the bill

    Finance, with every difference shown

  6. Person decides

    Pay the supplier

    Payments always stop at a person

Order-to-cash

From an enquiry to the money in. Agents read the enquiry, price the quote and draft the reminders; people send every message that reaches a customer.

  1. System

    An enquiry arrives

    Email, PDF or a photo from a phone

  2. Agent prepares

    Draft the quotation

    Document intake and quotation agents, priced from your lists

  3. Person decides

    Approve and send the quote

    Salesperson; sales manager below your margin floor

  4. System

    Deliver and invoice

    Your usual order, delivery and invoice steps

  5. Agent prepares

    Draft payment reminders

    Collections agent, when invoices pass due

  6. Person decides

    Approve and send reminders

    Credit controller

Works with what you have

Start on top of the tools you run today

You do not have to replace your accounting software or your inboxes to start. An agent can read from the tools you already use and put its drafts in front of your team. Which tools, and how the link works, is the first thing we check in the scoping call.

What you run today

  • Accounting softwareLedgers, invoices, bills
  • Shared inboxesOrders, invoices, requests
  • Spreadsheets and exportsPrice lists, stock counts
  • An ERP you already runRecords you want to keep

Agent

Reads what it needs, drafts the work

Read access only to the data its job needs. Its drafts wait in one approval queue.

Your team

Approves, edits or rejects

Where approved changes go, and how, is agreed with you and tested before go-live.

Want the whole system instead? The same agents sit inside the AI ERP we build, module by module.

Governance

Logged in your system, run by your rules

Agents are part of the system we build for you, so their limits, their log and their code are yours.

  • Logs in your system

    Every draft, edit, approval and rejection is recorded with who, what and when, in the system you own.

  • Least access per agent

    Each agent reads only the records its job needs, and writes only drafts.

  • Limits you set

    A control level per agent and per action, S$ thresholds, and second approvers where you want them.

  • An off switch per agent

    Switch any agent off and the work goes back to your team's manual path. The log stays.

  • Tested before go-live

    Agreed test cases, run by our tester, with the business rules checked by our ERP controller and every release reviewed by our founder. You sign off.

  • Model and data terms agreed up front

    Which AI model is used, what data it sees and on what terms are agreed with you in the scoping call and written into the scope.

Action log

Kept in your own system

  1. 07:01Purchase order agentDraftedPO-1187New draft, S$4,915.90 incl. GST
  2. 08:45Operations managerEditedPO-1187Bolts 300 → 250 boxes
  3. 08:45Operations managerApprovedPO-1187Draft → approved, S$4,240.10
  4. 08:45SystemSentPO-1187Emailed to supplier S-207
  5. 10:05AP invoice agentHeldINV-55021S$30.00 over PO, tolerance S$20
  6. 11:40Finance managerApprovedINV-55021Held → posted, S$1,166.30

Sample data. Harbourline Supplies Pte Ltd is not a real company.

Singapore guidance

Designed with IMDA's agentic AI framework in mind

Singapore's Model AI Governance Framework for Agentic AI, released on 22 Jan 2026, calls for significant checkpoints at which human approval is required. Our default of drafting for a person to approve was designed with that in mind. It is a framework, not a certification, and we do not claim to be certified under it.

Read the announcementSecurity and AI controls

Price

Start with one agent workflow

One agent workflow on your data as a quick build, from S$15k. You own the code, and there are no per-user licence fees.

Quick build

From S$15k

2 to 3 days of build. One agent workflow, on your data.

  • One agent, with its trigger, access and limits
  • The approval step and the action log
  • Tests on agreed cases, signed off by you
  • The source code, which is yours
  1. Step 1Scoping call
  2. Step 22 to 3 days of build
  3. Step 3Tests and your sign-off
  4. Step 4Go live, with the code
Book a scoping call See all pricing

  • Build

    S$30k to S$40k

    A set of modules and agents around one part of the business. 4 to 5 days of build.

  • Enterprise programme

    Around S$150k

    A full ERP programme, delivered phase by phase. Months, in phases.

  • After go-live: Run and Evolve

    Run covers hosting, support and AI usage. Evolve is monthly build capacity. Both are priced in the phase plan.

Looking at a grant?

EDG, PSG and MRA closed to new applications on 29 Sep 2026. The EDGE Grant applies from 30 Sep 2026. We never promise a grant outcome or percentage for a project. Read our EDGE guide.

FAQ

Questions about AI agents

What happens when an agent is wrong?

+
By default, agents draft and a person approves, so a wrong draft is caught before anything changes. The approver can edit or reject it, and the log keeps the draft, the change and the reason. Payments, deletions and messages to customers or suppliers always stop at a person. If an agent keeps getting one kind of case wrong, that case goes into its test plan, and the fix goes through the same tests and review as any release.

Which AI models are used, and is our data used for training?

+
We agree the model and the data terms with you in the scoping call, and write them into the scope before the build starts. That covers which model the agents use, what data is sent to it and on what terms. We do not quote a model provider's terms on this page; we go through the current terms with you.

What does an agent cost to run?

+
There are two parts. The build: one agent workflow on your data is a quick build, from S$15k. Running it: hosting, support and AI usage after go-live sit in the Run retainer, priced in the phase plan. AI usage depends on how many documents and records the agent handles, so we estimate it from your volumes. There are no per-user licence fees.

Can agents send email or WhatsApp messages?

+
By default, only after a person approves. Messages to customers or suppliers always stop at a person, whatever control level you choose. Email is the usual channel. Any other channel, such as WhatsApp, is scoped with you in the scoping call.

Can we switch an agent off?

+
Yes. Each agent can be switched off on its own. The work goes back to your team's manual path, and the log keeps everything the agent drafted and everything people approved.

Do we have to replace our current system?

+
No. Agents can work on top of the tools you run today, such as your accounting software and shared inboxes. Which tools, and what connecting to them needs, is the first thing we check in the scoping call. If you later want the whole system, the same agents sit inside the modules we build.

Who builds and checks the agents?

+
Three people. Lester Law, our founder, runs the AI coding agents that write the code and reviews every release. Our tester runs the test plan. Our ERP controller checks the business rules, such as prices, GST and who approves what. We are TeamSleek, the AI ERP team at SleekDigital in Singapore.

How long does one agent take to build?

+
One agent workflow is a quick build: 2 to 3 days of build once the scope is agreed. Before it goes live, it is tested on the cases we agree with you, and you sign it off.

Pick the agent your team needs first.

Bring one workflow that hurts. In 30 minutes we show you how it would run as an approval queue on sample data, and what a quick build would cost.

Book a scoping call See pricing