Skip to content
ProvidenceCrafts

Custom AI agents

The work, done by an agent that knows your business.

Every agent is built for one specific process in your business: connected to your systems, measured from the pilot and delivered as a product, with its own panel and its code.

Start the conversation

The cases we see most.

They are examples of what an agent looks like, not a catalogue: yours is built on your process.

Customer support

The same twenty questions, answered first time and at any hour.

An agent that answers with your real catalogue, prices and terms, qualifies whoever writes in and books time with your team when a person is needed.

Can include
Web, WhatsApp and emailLead qualificationBooking with your teamCRM integration
Still a person’s job
A delicate complaint, a negotiation or an angry customer: the agent knows to hand them to your team.

Documents and ERP

The delivery note stops going through a keyboard.

An agent that reads delivery notes, invoices and orders, turns them into data and puts them into your ERP, with the line that does not match held for a person.

Can include
PDFs, scans and photosMatching against orders and stockReview trayERP integration
Still a person’s job
An illegible document, a line that does not match the order or a new supplier: they stop for someone to review.
Delivery notes to the ERP. Self-initiated concept — document to structured data, with the line that does not match held for a person

Internal knowledge

Everything your company knows, askable in one sentence.

An internal assistant that answers from your manuals, procedures and technical documentation, and cites where it got the answer so it can be checked.

Can include
Manuals and proceduresAnswers with their sourcePermissions by person and areaModels on your infrastructure
Still a person’s job
Deciding which documents are the good ones: if one is out of date, the assistant cites it just as confidently. That is why it always cites the source.
Internal assistant on Open WebUI. Self-initiated concept — RAG over a (fictional) company’s internal documentation

Yours is different?

That is normal. Invoices, scheduling, reports, the inbox: if it repeats and can be measured, it can be built.

Start the conversation
  • Connected to your systems
  • Handover to a person
  • Its own panel and interface
  • Measured on your data
  • Models on your infrastructure or by API
  • Code handed over and documented

A generic tool makes you work its way. A custom agent works yours.

Today

With us

A tool made for any company

An agent built on your process

A demo that impresses on sample data

A pilot measured on your real data

Copying and pasting between the AI and your systems

Connected to your ERP, your CRM or your email

A flow only its author understands

A product with its panel and its code

Installed and left alone

We run it and adjust it every month

Not everything crosses to the right: a delicate complaint, a document that does not match or a decision with consequences still belong to a person, and the agent knows to hand them over. What stays in human hands is decided before building it, not when it fails.

Start the conversation
  1. Step 01

    Diagnosis

    Free. We measure the process on a real sample — conversations, documents, whatever goes through it: how much it repeats, what it costs and what has to stay with a person. If an agent does not pay off, we tell you.

  2. Step 02

    Proposal

    What the agent does, which systems it connects to and what goes to a person, with a fixed price before we start.

  3. Step 03

    Pilot

    Part of the process running on your data and measured against how it is done today, before opening it up fully.

  4. Step 04

    Operation

    It is connected all the way, and every month we review where it fails and adjust it. An agent improves with use; it is not left alone.

Before you write to us.

The questions that nearly always come up before we start.

A traditional chatbot follows a decision tree and does not understand anything it was not set up for. ChatGPT understands, but it does not know your data or connect to your systems, and someone has to paste text into it by hand. A custom agent reads the question or the document as it arrives, works with your real information inside your systems and knows when it does not know: at that point it hands over to a person. If your case is solved with ChatGPT, we tell you in the diagnosis.

That is the real risk, and it is designed against rather than hidden. The agent only works with what is in your information, cites where each answer comes from and marks its confidence; what it cannot find or does not match it does not fill in: it stops it for a person to see. And every month we review a real sample to see where it fails. None is always right; the difference is that you can tell when it does not know.

No. A flow chained together in an automation tool works for a demo and breaks when a screen changes or a reply arrives differently. What we deliver is a product: the agent, its panel so you can see and correct what it does, its integrations and its code. The code is yours and any developer can carry it on.

If your ERP, CRM or email accept data by API, file import or database, yes — and almost all accept one of the three. We check it in the diagnosis, before committing to anything. If one of them is a closed one, we tell you and see whether the alternative is worth it.

Wherever you decide, and it is decided before any code is written. We can work with models that run on your own infrastructure and never leave it, or with providers whose API terms exclude training on what they receive. Which one fits depends on the data: with invoices or internal documentation, the first is usually the sensible choice.

No. The diagnosis tells you which part of your data and documents is in shape and which part would confuse the agent: old versions, documents that contradict each other, processes nobody has written down. We start with what is good and sort out the rest by priority, not all at once.

Yes: the assistant in the corner of this page is a support agent like the ones we build. It knows the studio, how we work and our terms, and when a conversation needs a person it takes you to the form to start the diagnosis with us. Ask it anything.

Still got one? Write to us and we will settle it before you commit to anything.

Tell us where it gets stuck.

You get a reply within 24h, from the person who would actually build it. Four details, no sales sequence.

Prefer to write first? hello@providencecrafts.com

Ask for your free diagnosis

Four details and we will write back to set the first session.

The diagnosis is free: you pay nothing until you accept a proposal.