Skip to content
Comans Services
Menu

AI systems and agents

AI that fits the way your business works.

Less time entering information. More help getting things done. We build and improve AI systems for teams across Australia, connecting processes, applications and data.

Talk about your AI projectSee it in practice
Connected to the work
Processes, information and applications connect to AI working with the team, supporting capture, finding information and action.
  • Processes
  • Information
  • Applications
AI working with your teamConnected systems. Useful context.
  • Capture
  • Find
  • Act
Built around your rules and the work at hand.

Make everyday work easier.

Start with one useful change.

Capture as you work

Turn varied inputs into usable business records.

Find what matters

Put company knowledge and operational information to use.

Improve what you have

Refine an existing system’s quality, speed and running costs.

AI, put to work

Already helping people work.

Practical systems, connected to the applications people rely on.

A startup · In use

Timesheets, without breaking your flow.

For a startup working with AI tools, we built a timesheet system that lets an agent help capture work as it happens.

The agent handles varied inputs. Middleware helps check and structure the information before it reaches the application’s API, making time capture part of the work already underway.

We also use agents for CRM and timesheets in our own business.

From work to a usable record
Work in AI tools and varied inputs pass to an agent, through checks and clarifications, and through an application API into a timesheet record.
  • Work in AI tools
  • Varied inputs
  1. AgentPrepares the request
  2. Checks and clarificationsData quality · Business rules · Access
  3. Application API
  4. Timesheet
Checks are shaped around the workflow.

Intelligent monitoring · In use

Make sense of the signals.

System logs hold useful information, but someone still needs to work out what matters and what to do next.

Our live monitoring service uses AI to interpret logs, classify issues and create tickets with clear summaries and recommended next steps.

Model routing matches the processing to the task, with more capable models used where needed.

From signals to a clear next step
AI interprets system logs and prepares a ticket with a summary and suggested next steps for the responding team.
  1. System logsOperational events and signals
  2. AI interpretationUnderstand the issue and severity
  3. A ticket with contextA plain-language summary and suggested next steps
  4. Team responseInformation ready to act on

Company knowledge and local AI

Built around your information.

Useful answers start with the right context. We design AI around your sources, project access and operating needs, with local processing and cloud options where appropriate.

A local-first architecture
Access follows the person and project
Example local-first architecture: authenticated team access, project sources and memory feed relevant context, local AI processing and a response with supporting sources. Optional web research and approved cloud-model workloads are separate external routes.
Your teamAuthenticated access
Your managed environment
  • Project sources
  • Useful memory
  1. Relevant contextRetrieve and assemble
  2. Local AI processingRoute to a suitable model
  3. A useful responseSupporting sources
  • Optional web research
  • Cloud models for approved workloads
See the engineering layers
Identity and context

Entra ID, project access, retrieval, reranking and memory.

Models and infrastructure

Model gateway, routing, load balancing and GPU inference.

Ongoing visibility

Evaluation, usage tracking, running costs and operational logs.

Keep improving

Better results. Thoughtful use of resources.

We test the work that matters, match models to the task and refine the information they receive. Quality, response time and cost are considered together.

  1. Quality you can assess

    Check real examples, review exceptions and see where answers or actions need improvement.

  2. Models chosen for the task

    Use a suitable model first, with a controlled path to more capability when the work needs it.

  3. Running costs you can see

    Track usage and useful outcomes, then tune context, retrieval and model choices.

How we get there

From a useful idea to everyday use.

Build something new, or improve the AI you already use.

  1. Understand

    Find the friction, map the process and agree what a useful result looks like.

  2. Prove

    Test a focused approach with representative information and clear checks.

  3. Connect

    Build the integrations and controls, then introduce the system with your team.

  4. Improve

    Support it in use and refine quality, speed and cost as the work develops.

What we’re developing

Prism.

Make scattered information useful.

Prism turns mixed documents, emails, attachments and archives into structured information for AI workflows.

We’re developing it to make that information easier to organise, retrieve and reuse, while keeping a clear connection to the original source. It brings the work of preparing information closer to the agents and tools that use it.

Ongoing development
From mixed files to useful information
Prism converts mixed documents, emails and archives into reusable information with source references and linked attachments for retrieval into AI workflows.
  • Documents
  • Emails
  • Archives
  1. PrismConvert and structure
  2. Reusable informationSource references · Linked attachments
  3. Your AI workflowsRetrieve the material you need
Files are converted and stored locally. When you use a cloud AI provider, retrieved content is sent to that provider.

Electronics · Data · AI

AVeMe.

Electronics, data and AI, designed together.

We’re developing a wearable and local AI system that explores how personal context can make digital assistance more useful.

The work brings together low-power electronics, data systems and a small model intended to learn patterns over time, alongside a more capable local agent.

In development
The intended architecture
Intended AVeMe architecture, in development: a wearable feeds a phone and timeline, a personal model is intended to learn patterns, and a local agent supports the user. Experience and feedback inform the learning cycle.
  1. WearableSense
  2. Phone & timelineRemember
  3. Personal modelLearn patterns
  4. Local agentSupport
Experience and feedback inform the learning cycle.

Start a conversation

What would you like AI to help with?

Bring us a process, a promising idea or a system that could work better.

Talk to usOr call 1300 112 872