Skip to content
Agent frameworks · Australia

LangChain, LangGraph and CrewAI developer in Australia

Code-based agents in your own repository, with limits and a log of what they did.

LangChain, LangGraph and CrewAI are code libraries for building AI agents. A business may ask for them by name because a developer suggested one, or because a first test worked and now needs to become something a team can rely on. True North Innovations is a Melbourne-based team that works with businesses across Australia. We build agents with these libraries in your own code repository and your own cloud account, with written limits, a log and a point where a person approves. As a LangChain developer and LangGraph developer for a defined job, we handle LangChain development, LangChain implementation and LangChain integration with your own systems. We also tell you honestly when a job needs no framework at all. Our page on AI agent development explains agents in general.

Review · Build in your repository · Limits and logs · Written hand-over

Melbourne based.
Australia-wide delivery.

Melbourne · Sydney · Brisbane · Perth · Adelaide · Canberra

The problem

Where an agent that worked in a test goes wrong

01

A test that nobody can run again

The first version runs on one laptop, with keys typed in by hand and no record of how it was set up.

02

An agent with no limits

It can call any tool, use any file and run for as long as it likes, and nobody can say what it did.

03

A choice made too early

A framework was picked before the task was clear, and now the code is harder than the job needs.

Deliverables

What we build and hand over

01

An agent for one clear task

Built with the library that suits the task, in your own repository, with tests we run before you rely on it.

02

Written limits

Which tools it may use, which data it may read and what it must never do on its own.

03

Approval points

A person approves anything that changes a record, spends money or reaches a customer.

04

A log of each run

What the agent was asked, which tools it called and what it returned, so a person can check it later.

05

Settings kept apart from the code

Keys and passwords sit in your own secret store, never in the code or in a shared file.

06

Hand-over and training

We explain the code and the log to your developers, and write down who owns what.

Examples to assess. The scope is agreed in writing before work starts.

Right choice

When a framework fits and when it does not

The task is a fixed series of steps
How we see itA workflow tool is usually simpler to build and to hand over. See n8n or workflow and business process automation.
The agent must choose between steps by itself
How we see itA framework can help, with limits and a person approving. See AI agent development.
The data is special or the result must sit inside your own system
How we see itA custom build may fit. See custom AI development.
You only want answers from your own documents
How we see itA private AI assistant may be enough.
You are not sure an agent is the right answer
How we see itSay so. We would rather tell you at the start, and AI consulting can help you decide.

Many jobs do not need a framework. We say so when that is the case.

How it works

A worked example: an agent that prepares a supplier summary

A made-up example. An agent reads a supplier email, looks up the order, prepares a summary, a person approves and the note is saved.

Demonstration with dummy data

  1. 01 · Trigger

    A supplier email arrives

    A message reaches a shared mailbox and the run starts.

  2. 02 · AI step

    The agent reads and looks up

    It reads the email and uses one allowed tool to find the matching order.

  3. 03 · Check by a person

    A person approves

    The agent shows its summary and its proposed note. Nothing is changed until a named person approves.

  4. 04 · Action

    The note is added

    The approved note is added to the order record.

  5. 05 · Record

    The run is logged

    The email, the tool calls and the decision are kept in the log.

One agent run in five steps: the same five marks as every workflow on this website.
Errors and limits

Wrong answers, runaway runs and library changes

The agent gives a wrong answer
What we set upThe log shows what it used. We change the instructions or the data and test again.
A run goes on too long or repeats itself
What we set upA limit on steps and on time stops it, and a named person is told.
A tool call fails
What we set upThe agent stops and reports the failure. It does not guess.
The library changes with a new version
What we set upThe versions are fixed in the code, and an update is a planned change with a test run.
Cost of the AI service
What we set upUsage is charged by the AI service you choose, paid by you to its maker. We set limits on use.
What we need from you

What you give us to start

  • Access to your code repository and cloud account, set up so you can remove it later.
  • The task the agent should do, and what goes wrong today.
  • A list of the systems and data the agent may use, and the ones it may not.
  • Five real examples of the task, including an awkward one.
  • A named person who owns the agent after hand-over.
How we work

A clear process: scope, build small, test, hand over

  1. 01

    Scope

    We agree the task, the limits and where a person approves.

    Output: A short written plan

  2. 02

    Build small

    We build a small working version in your repository, using the library that suits the task.

    Output: A first working version

  3. 03

    Test

    We run it on your real examples and fix what we find.

    Output: Test results and a fix list

  4. 04

    Hand over

    We explain the code and the log and agree who owns the agent.

    Output: Training and written responsibilities

Cost

What affects the cost of agent framework work

How many tools the agent uses
Why it mattersEach tool needs access, limits and tests.
How clear the task is
Why it mattersA task that is still changing takes more rounds of testing.
How much checking is needed
Why it mattersMore approval points mean more design.
How well your data is organised
Why it mattersMessy data needs tidying before an agent can use it.
The AI service and your cloud account
Why it mattersThese are paid by you to their makers. They are separate from our work.

Every project is scoped individually.

Common questions

LangChain, LangGraph and agent frameworks: good to know

What can we use in place of LangChain?

It depends on the task. Options include other agent libraries, a workflow tool such as n8n or plain code. We choose after we understand the job, and we tell you in plain words why. Sometimes the answer is no framework.

Can you take over a prototype our developer built?

Often yes. We read the code, list what it does and what is missing, and agree the changes before we build.

Where does the code live?

In your own repository and your own cloud account. We do not keep a copy that you cannot reach.

Can the agent act without a person?

Only for steps you allow and that cannot do harm, such as reading. Anything that changes a record or reaches a customer waits for approval.

Will the agent always be right?

No. An agent can be wrong. That is why we set limits, keep a log and put a person at the points that matter.

Can you set up LangGraph or CrewAI in our own cloud account?

Yes. We build in your own repository and cloud account. We need a clear task, access you can remove later and a few real examples, and we agree the scope in writing before we build.

Explore more

Related AI automation pages and services

AI Automation (the whole subject) →

All our AI automation services, workflows, platforms and guides in one place.

AI agent development →

When the tool must take several steps on its own, within limits.

n8n consultant →

For flows that need more control and branching.

Custom AI development →

A small working tool built around your data when nothing ready-made fits.

A private AI assistant on your own documents →

Answers from your own files, with the source shown.

Agent, chatbot or workflow: which one? →

The difference between the kinds of AI help, and when each one fits.

Workflow and business process automation →

For whole processes with several people and approvals.

AI Consulting →

Not sure yet which tool or approach fits? Start with the advice.

See all services

Tell us about your business

Tell us about your agent project

Describe the task and what exists so far. We will tell you the clearest next step, in plain English.

Call 03 4158 4444info@truenorthinnovations.com.au

AI automation work is scoped individually.

We use your details to respond to your enquiry. Read our Privacy Policy.