MILabs
Tools

Lunary setup for small business AI monitoring

Lunary for small businesses: how MILabs sets up AI monitoring, what it records and what it fits.

Updated 2026-07-20

What Lunary is

Lunary is a monitoring and analysis tool for applications that call large language models. Once an AI feature is live, a business wants to know what it was asked, what it answered, where it failed and what each step cost. Lunary gathers those records into one place, so a team can review them without reading raw logs. Self-hosting is possible, which lets a business run it on infrastructure it controls. Teams use tools of this kind to follow conversations, trace the path a request took through several steps and check whether an assistant still behaves as its owner intended after a change. Lunary works alongside the model and the application, and it sits on the monitoring side of the stack. For a small business, its value is visibility, a clear and searchable record of what an AI feature did, so that later review and repair take less effort.

Lunary: official site

Lunary in plain language

When a business puts a language model behind a chat window, a quote form or a scheduled task, the output stops being a single answer that someone reads and approves. It becomes a steady flow of requests and replies, many of them produced while nobody is watching. Observability is the discipline of recording that flow so it can be read, searched and questioned later.

A single trace from such a flow holds the prompt that was sent, the context attached to it, the reply that came back, any tools the model called along the way, and the time and tokens each step consumed. Read together, these details show whether a failure came from the question, the retrieved material, the instructions or the model itself.

Hosting choices differ. Some tools are offered as hosted services, while others can run on a server the business manages, which matters when records contain customer names, quotes or health information. Teams also use the stored records to build evaluation sets, a fixed list of questions run against each new version of a prompt so that changes can be compared before customers see them.

What businesses use it for

  • Reviewing conversations between customers and an AI assistant to find answers that need correcting
  • Keeping a record of model usage so the owner can see where AI spending goes
  • Finding the step where an automated workflow failed when a request goes wrong
  • Comparing how an assistant responds before and after a change to its instructions
  • Checking that staff-facing AI tools give answers that match company policy
  • Sharing a readable log of AI activity with a team lead or an outside adviser

Who it suits

  • Owners with a customer-facing AI assistant who want to see its answers
  • Teams running several automated steps that call a language model
  • Technical leads who need a self-hosted record of AI activity

How MILabs sets it up

MILabs starts with the owner, choosing how Lunary will run for this business, either self-hosted on infrastructure the business already uses or through a hosted option. Together we decide which parts of the AI activity to record. We then connect Lunary to the applications and assistants the business already has, so calls to the language model are captured without rebuilding the product around them. Access is set by role, so only people who need the records can open them. We also agree who looks after the installation, who receives alerts when something breaks and who changes prompts or settings. That last point matters most, because a monitoring tool is only useful when someone is responsible for reading what it shows.

Builds it fits

Industries it fits

Lunary in practice

In a business that runs an AI feature, monitoring usually begins with a list of every place a model is called. That might be a website chat, a quote form or a scheduled job that summarises supplier emails. Each call point gets a clear name, so records from different parts of the business can be filtered and compared.

Day to day, the work is mostly reading. When a pattern appears, such as the same question failing after a policy document changed, the usual response is to revise the instructions or source material and then retest older cases to confirm nothing else broke.

Cost tracking forms a second strand. Usage records can reveal which assistant consumes the most tokens and whether an automated workflow calls the model more often than its purpose requires. The owner can then decide whether to shorten prompts, reuse stable answers or hand a simple sorting step to a smaller model. The records supply the evidence for each choice.

Access forms the third strand. When an outside adviser needs to look at the activity, the business decides what that person sees and removes customer details first.

good questions, straight answers

Frequently asked questions

Can my team use Lunary without training?
Yes, with a short introduction to each view, which MILabs can walk through as part of setup. Lunary suits people who review AI activity. A team that already uses its AI assistant day to day usually needs less guidance than one that has never looked at model output. MILabs can focus on the views each person will use most often.
Can MILabs build on top of Lunary?
Yes, MILabs can connect Lunary to the applications a business already runs and build reviews, summaries or alerts around the records it collects. Lunary is the monitoring layer, so the assistant or workflow usually comes first, with Lunary added to watch it. Where a business needs a custom view of its AI activity, that work can be scoped through the enquiry form. Any build is agreed with the owner before work starts.
What does a Lunary setup need from me?
It needs an owner or team lead who will decide what is recorded, who can see it and how it is hosted. It needs access to the applications and assistants that call a language model. Someone must be named to read the records and act on what they show, because a monitoring tool does nothing on its own. MILabs will ask for these details at the start and confirm them in writing first.
What does Lunary not do?
Lunary cannot make an AI assistant more accurate by itself, and it cannot replace testing or human review of what an assistant says. It records and analyses activity after an application has made its calls, so it cannot fix a weak instruction or a missing document on its own. Deciding what a business should automate stays with the owner. Its job is to make an AI feature's behaviour visible.
Is Lunary suitable for a small business?
Yes, for a small business that has a live AI feature and someone willing to review its records. A business with no AI activity has little for it to watch, and one with no one to read the output will find it hard to get value. Self-hosting suits owners who want their records kept on infrastructure they control. MILabs can advise on whether it fits a business through the enquiry form.
let's make something happen

Workflow automation: how we can help

Choose how you want to work on this with MILabs. Both options start with an enquiry.

Discovery is free, no obligation from the first chat.

Let us build it for you

We build it and look after it.

Build it yourself, we'll help

You build it with our direction, teaching and a weekly check-in.