AurelithCustom AI agents with a real job and clear limits.
In short
Aurelith designs and builds custom AI agents for operating companies in Greece and Europe. Each agent takes on one defined responsibility, uses only the context and tools you approve, keeps the work moving and hands the decision to a person when judgement is needed. Agents are part of Level 02, Custom AI Solutions.
- Free first conversation
- Proposal before any commitment
- Support with no end date
Business first. Intelligence at every level.
What is an AI agent, and what is it not?
A chatbot answers. Traditional automation follows fixed rules. An agent moves a piece of work forward, and Aurelith builds one where the work changes from case to case.
An AI agent is software that moves a defined piece of work forward. It chooses the next step itself, within limits you set.
To do that, an AI agent uses an AI model to interpret the goal, retrieve the information it is allowed to see and call the tools it is allowed to use. It then finishes the task, or hands it to a person when judgement is needed. What makes an agent useful is not the model but the responsibility: a bounded job, the information and tools that job needs, rules for what it may do and a clear definition of done.
Aurelith builds custom agents: each one is designed around one responsibility in your business and connected to the systems your people already use. Where fixed rules are enough, Aurelith uses rules; the agent takes on the parts that need reading, interpretation or a flexible order of steps.
| Question | Chatbot | Traditional automation | Custom AI agent |
|---|---|---|---|
| What it does | Answers questions and drafts text inside a conversation | Runs the same sequence of rules every time | Moves one defined responsibility forward, step by step |
| Who chooses the next step | The person asking, at every turn | The rules, written in advance | The agent, within the tools and permissions it is given |
| Best for | Guidance, quick answers and first drafts | Stable, high-volume steps with known conditions | Multi-step work with documents, language and exceptions |
| Where people decide | On whatever they do with the answer | When an input breaks the rules | At the approval points you set, and wherever the agent reaches its limits |
What work can an AI agent take on?
Five kinds of responsibility that recur in operating companies. In each one the agent keeps the work moving, and a named person keeps the decisions that matter.
- 01
Document-heavy operations
The agentReads, classifies and extracts from contracts, forms, reports and correspondence, then prepares the next action.
A person keepsAnything that commits money, signs off a document or leaves the company.
- 02
Knowledge and research
The agentFinds relevant internal and external material, checks the sources and assembles the context a decision needs.
A person keepsThe decision itself, with the sources in front of them.
- 03
Operational coordination
The agentFollows status, owners, exceptions and next actions across teams and systems, and flags whatever is stuck.
A person keepsThe priorities, and any change to a plan or a commitment.
- 04
Service requests
The agentUnderstands each request, gathers the facts it needs, carries out the approved actions and escalates anything non-standard.
A person keepsThe exceptions, and every case outside your policy.
- 05
Commercial operations
The agentResearches and qualifies opportunities, prepares the follow-up and keeps CRM records current.
A person keepsWhich opportunities to pursue, and every message to a client.
What goes into an agent you can rely on?
Custom AI agent development at Aurelith starts with six decisions, settled before the agent touches live work. Together they define what it may see, what it may do and who answers for it.
- 01
A bounded responsibility
One job, a clear outcome, a definition of done and an explicit list of what stays out of scope.
- 02
Approved context
The documents, data, policies and records the agent may use, and how they are kept current and access-controlled.
- 03
Well-defined tools
Specific capabilities to read or to act, each with clear inputs and outputs and a known behaviour when something fails.
- 04
Permissions and guardrails
Limits on what the agent can see and do, approval checkpoints for higher-impact steps and escalation for anything out of bounds.
- 05
Evaluation on representative cases
Tested before launch on normal cases, edge cases and the ones that must escalate, then refined in real conditions with your team.
- 06
Human ownership
Named people who answer for the workflow, review how the agent behaves and decide when it should change or stop.
How does an agent earn more authority?
Every agent Aurelith builds starts with the least authority its job allows. It climbs the same ladder Company Brain uses, one action at a time, and only when the evidence says it is ready.
AI receives authority action by action, only when evidence shows it is reliable.
Granted by people · lost when the evidence weakens
A new agent usually works on the first three rungs: it observes, recommends or drafts, and a person decides. The two rungs above are earned later, or not at all: approval-bound execution only after explicit permission, and bounded autonomy only for proven, low-risk work inside strict limits.
Authority is granted action by action, not to the agent as a whole, and you decide every step up. At Level 03, Company Brain applies the same ladder to every workflow, action, tool, context and model in the company.
- R0Read-only
Observe
Read and organise authorised information.
- R1Human decides
Recommend
Suggest a next step without taking action.
- R2Approval required
Draft
Prepare an action for authorised review.
- R3Controlled action
Approval-bound execution
Execute only after explicit permission.
- R4Verified continuously
Bounded autonomy
Perform proven, low-risk work inside strict limits.
Go deeperHow Company Brain governs AI agents Guide: the autonomy ladder, rung by rung
Can an agent work inside your own software?
Yes. When a process needs its own screen, Aurelith builds it around the agent, so your people review, approve and follow the work in one place.
-
Custom interfaces
A screen designed for one process, where your team sees what the agent prepared and approves the next step.
-
Internal platforms
One place for a whole process: requests, documents, status and approvals, with the agent working inside it.
-
Customer portals
Where your clients submit and follow their requests, while the agent gathers the facts and your people keep the decisions.
Every interface connects to the systems you already use, such as email, documents, CRM and ERP, and belongs to Level 02, Custom AI Solutions.
Level 02 on ServicesAurelith also builds EnSign, ship registration software for flag registries and shipowners. The agents and interfaces in your project are built by the same company.
Visit EnSign (opens in a new tab)What does a custom AI agent project give you?
Every custom AI agent project with Aurelith delivers these six things. The exact scope and deliverables are set out in your written proposal, agreed before anything is built.
The payoff is momentum: the work keeps moving without anyone chasing it, and every decision that matters still reaches the person who owns it.
- 01
One agent, one responsibility
A defined job with a clear outcome, and only the context, tools and permissions that job needs.
- 02
The limits you agreed
Approval points for higher-impact steps, and escalation to a person for anything out of bounds.
- 03
Tested on representative cases
Normal cases, edge cases and the ones that must escalate, before the agent handles live work.
- 04
A view of its work
Your team can see what the agent read, prepared and did.
- 05
A team guided to work with it
Aurelith tests the agent in real conditions, refines it and guides the people who work alongside it.
- 06
Support with no end date
Aurelith stays with you after launch, so the agent keeps pace as your business changes.
From first conversation to a working agent.
Every Aurelith service follows the same six steps, and nothing is built before you agree the proposal.
First conversation
Free, and without commitment. You tell us which work you would hand to an agent, and we ask about the workflow, the systems and the people around it.
Analysis of your business
We look at your processes, your tools and how your people work, to find the responsibility an agent can carry well.
Proposal
In writing: service level, scope, deliverables, timeline and investment, agreed before any implementation starts.
Build together
We build the agent with your team, show the progress as we go and shape it with your feedback.
Put it to work
We test it in real conditions, refine it and guide the people who work alongside it.
Support with no end date
We stay with you after launch, so the agent keeps pace as your business changes.
Where do custom agents fit in the three levels?
Custom AI agents sit at Level 02, Custom AI Solutions, alongside AI automation. Aurelith usually starts with one agent; coordinating many agents across a company under one set of governance rules is what Company Brain does at Level 03. You can start at any level; the first conversation shows where.
The three levels, side by side, with what each one includes.
All AI servicesQuestions before your first agent.
Scope and pricing across all three levels are covered on Services. Anything else, ask us in a free first conversation.
What is a custom AI agent?
A custom AI agent is software built around one responsibility in your business. It uses an AI model to choose the next step, works only with the context and tools you approve, and hands the decision to a person when judgement is needed. Custom means it is designed around your workflow, your rules and your systems, not configured once for everyone.
How does Aurelith decide between an AI agent and a chatbot?
Aurelith recommends a chatbot when people mainly need answers, guidance or drafts inside a conversation, and an agent when the work has to move forward across steps and systems: reading a request, gathering the facts, preparing or taking a permitted action and escalating whatever it cannot settle. It makes that call once it understands the workflow, and the two often work together. The guide AI agents for business compares chatbots, traditional automation and AI agents.
Can a custom agent work with our CRM, ERP and email?
Yes, where your systems allow secure access. Aurelith connects each agent to the tools your people already use, such as email, documents, CRM and ERP, with only the read and write permissions its responsibility needs. Anything the agent writes back follows the approval points you agreed, and your systems remain the record.
Who decides what the agent is allowed to do?
You do. Before the agent handles real work, Aurelith agrees its responsibility, permissions and approval points with you, and your team can see what it read, prepared and did. A new agent typically starts by observing, recommending or drafting for approval, and it receives more authority one action at a time, only when evidence shows it is reliable and you decide to grant it.
How does an AI agent project with Aurelith start?
With a free first conversation about one valuable workflow. Aurelith then analyses how the work runs today and sets out the scope, deliverables, timeline and investment in a written proposal. Once you agree it, the agent is built with your team and tested on representative real cases, including the ones that must escalate, before its role grows.
What happens after the agent goes live?
Aurelith tests the agent in real conditions with your team, refines it and guides the people who work alongside it, then stays by your side with support that has no end date. Its authority grows only one action at a time, when evidence shows it is reliable and you decide to grant it.
What does it cost to build a custom AI agent?
Aurelith has no public price list. The first conversation is free, and the investment is set out in your written proposal once the work is understood. For an agent it depends mainly on the number of systems it connects to, how varied and exception-heavy the work is, how the approvals are designed, whether your people need their own interface and the quality of the data it relies on. More on scope and pricing.
AI agents / Start here
Give an agent one real responsibility.
Start with a free first conversation about one workflow. If an agent is the right answer, you will see its responsibility, its limits and the investment in a written proposal before anything is built.
No public price list. Scope, timeline and investment are set out in your proposal.