01
Custom AI agents
Task-specific agents with defined tools, boundaries and escalation rules — built for one job and measured on how well they do it.
AI agents · Custom AI software
AI that does the work, not just the demo.
METALCORE GLOBAL designs and builds custom AI agents and AI software for real operations — customer service, quoting, order handling and reporting — deployed in your environment, with human approval wherever a decision carries weight.
support
quoting
ops
Intakeenquiry read into structured fields
Quote draftpricing rules applied from your list
Approval gateheld for a named approver
CRM syncrecord and follow-up written back
Confirmationheld until approval returns
Illustrative configuration — not client data.
What we build
Six systems we build again and again.
Each one can stand alone or combine into a single operating layer. We start from the workflow that costs you the most hours, not from a technology list.
02
AI customer service
Multilingual first-line support that answers from your own documentation, qualifies the request, and hands off the high-value cases to your team.
03
Quoting and RFQ automation
Inbound enquiries are parsed into structured requirements, matched against your authorised pricing, and routed for approval before anything is sent.
04
Order and back-office automation
Documents extracted, orders structured, confirmations and reminders sent — the repetitive work that quietly consumes a week each month.
05
Analytics and business reporting
Enquiry, opportunity and order data turned into daily, weekly and monthly reporting — with the signals that actually change a decision.
06
Integration and handover
Connection to your CRM, ERP, email and payment systems through documented APIs — plus the runbooks your team needs to own it afterwards.
How we work
Accuracy over enthusiasm.
An AI system that is wrong 10% of the time is not a 90% win — it is a new review burden. We build for the accuracy your operation can actually absorb.
-
Scope before code We write down what the system does, what it must never do, and how success is measured — before development starts.
-
Approval where it matters Quotes, payments and order confirmations can be routed to a named approver instead of being sent automatically.
-
Your accounts, your data Models are called from infrastructure you control. Customer data is not used to train third-party models.
Delivery shape
Five stages, each one payable on its own.
You are never asked to commit to a full platform up front. Stage one produces something you can test against real cases.
01
Discovery and prototype
Business model, users, data sources, approval rules — plus a working prototype of the core interaction.
02
MVP
One agent or one workflow, running on your real inputs, with a management view.
03
Integration
CRM, ERP, email, documents and reporting connected through APIs.
04
Test and deploy
Accuracy, security, performance and cost tested before general use.
05
Scale
Additional agents and workflows added as usage and ROI justify them.
From enquiry to running system
What an engagement looks like.
Timelines below are typical for a first agent or workflow. Anything larger is quoted after discovery, not before.
1
Requirements and architecture
We map the current process, the data it depends on, and the decisions that must stay human. Output: solution architecture, interface specification and a fixed scope.
1 – 2 weeks
2
Prototype against real cases
The core interaction is built and tested against cases from your own history. You see the failure modes early, while they are still cheap to fix.
2 – 3 weeks
3
Build and integrate
The agent or application is developed, connected to your systems, and instrumented with logging, permissions and approval routing.
4 – 8 weeks
4
Evaluation and controlled rollout
Accuracy, latency, cost per task and security are measured. Rollout starts with a supervised group, then widens as the numbers hold.
2 – 3 weeks
5
Handover and support
Documentation, runbooks and access transfer to your team. Ongoing support is optional and separately priced.
Ongoing, optionalStraight answers
Six questions we get before the first call.
Short answers, with a link to the page that goes deeper. If your question is not here, ask it on the call — we would rather answer it early than have it decide the project later.
Are you a software company or an AI company?
We build software; the model is one component inside it.
We are a software company. The model is one component inside a system we specify, build, integrate and hand over. On our own pages that shows up as scope written before code, an evaluation set drawn from your real cases, and source code plus runbooks delivered at the end. If a model is the wrong tool for the job, an engineering company can say so — a model vendor cannot.
We already have a CRM/ERP — do we have to replace it?
No. We integrate with what already holds your data.
We connect to it. Our integration work goes through the documented APIs of your CRM, ERP, email and payment systems, because AI that cannot reach your systems is a chatbot. Nothing on our site asks you to migrate. If your data sits in a system with no usable API, we tell you that during discovery — before a quote exists, not after the build starts.
How long does one AI agent take to go live?
9–16 weeks of staged work for a first agent, plus handover.
One agent or one workflow usually runs through five stages: requirements and architecture in 1–2 weeks, a prototype against your own cases in 2–3 weeks, build and integration in 4–8 weeks, evaluation and controlled rollout in 2–3 weeks. Handover follows, and ongoing support is optional. Those ranges assume one agent and a known set of systems. Anything wider is quoted after discovery, not before.
Who is responsible when the model gets it wrong?
Your named approver decides. Our liability is capped at fees paid.
The decision does not leave your company. Every engagement names the approvers who authorise quotes, payments and order confirmations, and low-confidence cases go to a review queue instead of being sent. Our terms require those human review steps to stay in place. Contractually, our liability is limited to the fees paid for the engagement, so you should read the terms before signing.
Who owns the data and the code?
You own the deliverables; we license our pre-existing tooling to you.
You do — the deliverables, including source code and prompts written for your engagement, unless we agree otherwise in writing. We keep our pre-existing tools and templates and license them to you as embedded. Systems run in infrastructure you control or that is contracted in your name, access is logged, and client data is not used to train third-party models.
Why can't you just give me a price?
Four factors set the number; all four come out of discovery.
Because four things move the number, and we would be guessing on all four: the scope of the workflow, how many systems we have to integrate, how ready your data is, and how soon you need it. All four are established during discovery. What we can do now is quote a fixed-scope first stage, or run a paid two-to-three week prototype before anyone commits to a build.
Next step
Tell us which process costs you the most hours.
Send the workflow, the volume and the systems involved. We reply with an honest read on whether an AI agent is the right answer — and a scoped quote if it is.