Your rules, not the model's
Permissions and business rules stay in your application. The model only sees what the signed-in user is allowed to see.
AI development and integration, built into your application
We add AI features to Laravel and PHP applications, and build new products with AI at the core. Brisbane and Gold Coast team, working Australia-wide.
Request QuoteAI that earns its place in the product
Coded is an AI development company in Brisbane and on the Gold Coast, working with businesses across Australia. We build AI features into web applications: reading documents, answering questions from your own data, assistants for staff and customers, and agents that act inside your systems. Most of our work is in Laravel, so the AI lives in an application you own.
We are a good fit if you already run a Laravel or PHP application and want AI inside it, if you are a founder building a product with AI at its core, or if you are an agency that needs the AI part of a project built properly.
One example: DeckPress, a Laravel product we are building, uses AI to generate the custom artwork for every deck of playing cards it produces. Read how AI makes products like it possible.
What we build
Most of what businesses ask AI to do falls into six kinds of feature. Each one runs inside your application, behind the logins and permissions you already have.
-
Document processing
Invoices, forms and contracts read into structured data your app can use, with uncertain results flagged for a person to check.
-
Answers from your own data
Staff or customers ask questions of your manuals, policies or past tickets and get answers that show where they came from.
-
Assistants and chatbots
Grounded in your content, honest when they do not know, and able to hand the conversation to a person.
-
Sorting and routing
Enquiries, support tickets and leads read, summarised, tagged and sent to the right person.
-
Content tools in your admin
Drafts, summaries and translations inside your CMS or back office, with an approve step before anything goes live.
-
Agents and MCP servers
Let Claude, ChatGPT or your own agent use your application through a small set of tools, each with its own permissions.
Laravel at the centre, any model behind it
Calling an AI model is the easy part. Whether a feature holds up depends on everything around that call, and Laravel's queues, policies and tests are built for exactly that work.
- Slow model calls run on queues, so pages stay fast.
- More than one provider sits behind the same interface, so a price change or an outage is a configuration change.
- Results are cached and spending is capped from the first day.
- Your application decides who can see what. The model never does.
- Before launch we build a set of real examples with the answers you expect, and run it again whenever a prompt or model changes.
- Every call is logged with its cost, time and result.
- Laravel AI SDK
- Prism PHP
- Anthropic Claude
- OpenAI
- Google Gemini
- pgvector
- Laravel Horizon
- Pest
Your data stays under your control
An AI feature sends data to a model provider. We decide with you what is sent, where it goes and what it may be used for.
Only what the feature needs
We send the smallest amount of data that does the job, and leave out personal details the feature has no use for.
Training terms checked first
Before anything is sent, we go through the provider's data terms with you and choose a setup that keeps your data out of model training.
Australian hosting where it matters
If your data has to stay in Australia, we tell you which models can run here and which cannot, before anything is built.
Our Process
From first call to running in production
We prove the AI works on your data before it goes into the product, so the big decision is made on evidence rather than a demo.
-
Scope the job
What the AI has to do, the data it will use and what a good answer looks like, agreed in plain terms.
-
Prove it on your data
A small working version run against your real documents or records, so you see the quality and the running cost first.
-
Build it in
Into your application with queues, permissions, cost limits, tests and the example set that guards against regressions.
-
Run and improve
Monitoring, model updates and prompt changes after launch, by the same developers who built it.
What AI development costs
An AI feature has two costs. The first is the build: connecting the model to your data, checking its output and fitting it into your application. The second is the running cost, paid to the model provider each time the feature is used and usually measured in cents per document or conversation.
The build depends on how tidy your data is, how much checking the output needs and how far the feature reaches into your app. We quote after a scoping call, and for an existing application after we have opened the code. Running costs are estimated from the working version on your own data, so you know the figure before launch.
We build with AI tools every day. A senior developer owns every change.
Frequently Asked Questions
Connecting an AI model to your application so it can work on your data inside the workflows you already have: reading a document, answering a question, sorting a request. Most of the effort goes into what surrounds the model: getting it the right data, checking what comes back, handling failures and keeping costs predictable.
There is the build, and there is the running cost paid to the model provider for each use. The build depends on your data and how deeply the feature reaches into your app, and we quote it after a scoping call. Running costs are usually cents per document or conversation, and we measure them on your data before launch.
Yes. We start with the repo and the data, then add the feature behind your existing permissions so it fits the app rather than sitting beside it. If an older PHP application needs groundwork first, such as an upgrade or tests, we say so in the quote.
It depends on the job. A larger model suits reasoning over long documents; a smaller, cheaper one suits sorting and tagging. We build with more than one provider behind the same interface, test the options on your data, and keep it easy to switch when prices or quality change.
The major providers do not train on data sent through their paid business APIs by default, but terms differ between providers and change over time. We check the current terms for the provider we choose with you before any data is sent, and set the feature up to match.
The working version on your own data comes first and is the quickest part. The full build depends on how far the feature reaches into your application. You get a timeline with the quote.
Yes, when a chatbot is the right answer. Ours answer from your own content, say when they do not know, and hand the conversation to a person. For many businesses a search that answers questions, or a tool for staff, does more good than a chat window on the home page.
Yes. Agencies and studios can bring us in for the AI part of a project, and we hand the work back in a standard Laravel layout their own developers can maintain.