What does an AI agent cost?

Mapping is a fixed fee. Builds are priced as fixed phases once the scope is clear. Running costs such as hosting and model usage are estimated before you commit, capped, and reported every month. We share ballpark figures on the first call.

Applied AI · Colombo · Worldwide

Put AI on the payroll.

Q: What does XD Teknika do?

XD Teknika builds AI agents that take on specific, repetitive work inside a business: answering customers, following up leads, processing documents and finding answers in internal knowledge. We design them, connect them to your systems, launch them carefully and keep them running. We are based in Colombo, Sri Lanka, and work with teams worldwide.

Q: Why not buy an off-the-shelf AI tool?

If a tool fits your process, buy it, and we will tell you when that is the case. We earn our place where the work crosses several systems, follows your own rules and has to run unattended. We often build on top of software you already pay for.

Q: What happens when the agent gets something wrong?

Agents are built to recognise uncertainty. Unclear cases go to a person with the full conversation attached, risky actions need approval, and every mistake is reviewed and added to the test set so it does not repeat.

We design, deploy and run AI agents that answer customers, follow up leads and clear the back-office queue. Each one starts with a target you sign off and reports against it every month.

Book a 30-minute call Meet the agents

+You own the code
+Any model, no lock-in
+Humans approve what matters
+PDPA and GDPR aware

Plugs into what you already run

Your channels, your systems, and whichever model wins on your data.

WhatsAppGmailOutlookHubSpotSalesforceZohoShopifyWooCommerceZendeskXeroQuickBooksSlackMicrosoft TeamsGoogle Sheets

OpenAIAnthropic ClaudeGoogle GeminiMeta LlamaMistralDeepSeekModel Context ProtocolAWSAzureGoogle Cloud

01 / The agents

Start with the work that never stops

Most companies don't need an AI strategy. They need the inbox answered at 2 am, the lead called back in five minutes and the invoices keyed in by Friday.

Support Agent

Resolution rate

Answers customers on chat, WhatsApp, email and the phone, then does what they asked. It can reschedule a delivery, start a return or change a booking on the spot.

Explore the Support Agent

Sales Agent

Speed to lead

Replies to every enquiry within a minute, qualifies it against your criteria and books the meeting into the right calendar. Your reps spend their day with buyers.

Explore the Sales Agent

Operations Agent

Hours returned

Reads invoices, forms and emails, enters the data where it belongs and chases whatever is missing. A person signs off anything that moves money.

Explore the Operations Agent

Knowledge Agent

Answer accuracy

A private assistant that knows your policies, contracts and manuals. Every answer cites its source, and it can run inside your own cloud.

Explore the Knowledge Agent

Got a process that doesn't fit a box? We build custom agents too.

02 / Under the hood

11:47 pm. A customer sends a message.

Your team logged off hours ago. Here is what a Support Agent does in the next few seconds, and why you can trust every step.

WhatsApp · Customer

Hi, can I move order 4471 to Saturday? Also need another pair in size 42.

Done. Order 4471 now arrives Saturday between 9 and 12. Size 42 is in stock and I've reserved a pair for you. Your new total is LKR 18,400. Tap below to confirm and pay the difference.

Confirmed, thanks!

Order updated · courier rebooked · CRM logged

support-agent / tracelive

+0.00sRECEIVEWhatsApp message, customer matched by phone

+0.41sPLAN2 intents: reschedule_delivery, add_item

+0.88sTOOLorders.get(4471) ok · belongs to this customer

+1.32sTOOLcourier.slots(sat) ok · 09:00 to 12:00 open

+1.70sTOOLinventory.check(size 42) ok · 3 in stock

+2.05sPOLICYprice changes, customer must confirm first

+3.40sREPLYsent, waiting for confirmation

+41.2sACTorder updated, courier rebooked, CRM logged

FACTS

Answers come from your systems

Stock, slots and order history are looked up live. The model never guesses a delivery date.

RULES

Policy decides what it may do alone

Rescheduling is allowed. Changing what a customer pays needs their confirmation. Big refunds wait for your team.

RECORD

Every step is logged and replayable

If something looks off, you can see exactly what the agent saw, decided and did, down to the millisecond.

03 / How we deploy

Each step has to earn the next one

No twelve-month roadmaps. You see a working agent on your own data early, and you can stop after any phase with something useful in hand.

01 · MAP · 1 TO 2 WEEKS

Find the work worth automating

We sit with the people doing the job, pick one process and agree the number that defines success. The plan is yours to keep, whoever builds it.
02 · PROVE · 4 TO 6 WEEKS

Run it on real data

A working agent on one workflow, with a person checking its output. We measure before and after, and the number decides what happens next.
03 · LAUNCH · IN STAGES

Roll out with a safety net

Traffic moves over gradually, with fallbacks, alerts and training for the people who will work alongside the agent.
04 · RUN · MONTHLY

Keep it sharp

We watch quality and cost, turn mistakes into test cases and upgrade models as better ones ship. Or your own team takes the keys.

See how it works in detail

04 / Commitments

The fine print, in large type

AI projects tend to stall for the same few reasons. These are the rules we work by so yours doesn't.

You own everything

Code, prompts, workflows, test sets and documentation. Accounts sit in your name from day one.

One number per agent

Agreed before we build. If the pilot doesn't move it, we tell you and we stop.

People approve what matters

Refunds, discounts, payments and anything irreversible wait for a human, at thresholds you set.

Private by default

Your data never trains public models. When the data demands it, the agent runs inside your own cloud.

Any model, no lock-in

We choose models by testing them on your cases, and swap them when something better ships.

Costs you can forecast

Fixed-price phases. Model and hosting spend estimated up front, capped and reported monthly.

Read our trust standards

05 / Where we work

Built for businesses that run on conversations and paperwork

The agents stay the same. The rules, systems and vocabulary change by sector, and we learn yours first.

Retail and e-commerceOrder questions, returns and abandoned carts Hospitality and travelBookings, changes and pre-arrival questions Financial servicesClaims intake, document checks and leads Healthcare and clinicsAppointments, reminders and admin Logistics and distributionShipment status, PODs and invoice matching Professional servicesClient intake, research and knowledge search

06 / Questions

What people ask on the first call

Something we haven't covered? Send it over. A person on the team will reply within one working day.

Ask the team directly

What does XD Teknika do?

We build AI agents that take on specific, repetitive work inside a business: answering customers, following up leads, processing documents and finding answers in internal knowledge. We design them, connect them to your systems, launch them carefully and keep them running. We're based in Colombo, Sri Lanka, and work with teams worldwide.

Why not buy an off-the-shelf AI tool?

If a tool fits your process, buy it. We'll tell you when that's the case. We earn our place where the work crosses several systems, follows your own rules and has to run unattended. Often we build on top of software you already pay for.

How fast will we see something working?

Mapping usually takes one to two weeks. A pilot on one workflow typically runs on your real data within four to six weeks. Access to systems and data sets the pace, and we'll flag an unrealistic timeline in the first meeting.

What does an agent cost?

Mapping is a fixed fee. Builds are priced as fixed phases once the scope is clear. Running costs like hosting and model usage are estimated before you commit, capped, and reported every month. We share ballpark figures on the first call.

Is our data safe?

Your data is never used to train public models. Each agent can only reach the systems it needs, every action is logged, and sensitive workloads can run entirely inside your own cloud or servers. We design for Sri Lanka's Personal Data Protection Act and the GDPR. More on trust.

What happens when the agent gets something wrong?

Agents are built to recognise when they might be wrong. Unclear cases go to a person with the full conversation attached, risky actions need approval, and every mistake is reviewed and added to the test set so it doesn't repeat.

Do you work outside Sri Lanka?

Yes. We work in English with teams in the UK, Europe, the Gulf, Australia and Asia. Our time zone (UTC+5:30) overlaps the working day in all of them, and most engagements run remotely.

Your move

Bring us the job nobody wants.

Thirty minutes, one process, and a straight answer on whether an agent should take it on. No slides and no obligation.

Book a 30-minute call Describe the process