Mickai's free AI agents: what they do, and why they are free during beta

Trust Agent, the free AI education site published by Mickai LTD, is building a small set of downloadable AI assistant personas: named agents such as Marcus, a code reviewer, Raj, a project manager, and Ruth, a research assistant. Each one is a configuration file rather than a piece of software, a focused system prompt and a restricted toolset, written out in full so you can read exactly what it will do before you use it. They are free to download while the product is in beta. That is not a permanent promise, and the first set is still finishing review before it goes live. This is a plain account of what they are, how they differ from just asking a chatbot the same question, and where the free-during-beta decision actually stands.

What a Trust Agent persona actually is

Strip away the branding and an agent is a text file. It carries a name, a role, a description of what it is for, and a system prompt, usually a few hundred words, that tells whichever AI tool is reading it how to behave: what it knows, what it should refuse, and how it should talk. Trust Agent publishes that text in the formats four different tools expect: as a Claude Code subagent, as an OpenCode agent, as a VS Code custom agent (the format GitHub Copilot's tooling reads), and as a plain block of text you can paste into any chat tool that does not support a special format at all.

Marcus reviews code for security weaknesses, bugs and unclear design the way a careful senior engineer would, and says so plainly when something is outside a quick review. Raj organises a project: breaking work into tasks, flagging risk and keeping a plan honest. Ruth reads and summarises source material, distinguishing what a document actually says from what it might imply. Each one has a stated role and stays inside it; none of them claims to be a real person, a licensed professional, or an authority to defer to instead of your own judgement.

How this differs from asking a chatbot the same question

A general-purpose chatbot can do a passable job of the same request if you prompt it well. What an agent adds is not a smarter model underneath, it is usually the same underlying model you would use anyway, but three things a loose prompt does not give you: a fixed scope (Marcus reviews code; he does not also try to plan your sprint), a published instruction set you can actually read before you rely on it rather than trusting an unseen prompt buried inside a product, and a restricted toolset. By default, a Trust Agent persona can only read files, search text and list what is there; it cannot run commands, write or change files, or reach the network, unless the tool you run it in is separately configured to allow that. That default is a deliberate choice: a code reviewer that can only read code is far easier to trust than one that can also quietly change it.

Where things stand right now

Being honest about the current state matters more here than sounding finished. As this is written, Trust Agent's agent downloads are still in final review and are not yet available on the live site. Everything described above is the design, not something you can download today. Reviewing an agent properly, checking who owns the underlying material it was built from, what it is allowed to do, and whether its instructions or any supporting knowledge contain anything that should not be there, takes real time, and the decision has been to finish that review before publishing rather than ship first and fix problems afterwards. What is already live and free today is Trust Agent's separate library of AI course workbooks, unaffected by this and available now at trust-agent.ai.

Why free during beta, and why that is not a promise forever

The plan is straightforward, and not unusual: release the agents free while real people try them, see which ones turn out to be genuinely useful, and only then work out a pricing model that reflects what people actually value rather than a guess made before anyone had used the thing. That is normal product development, not a trick to build a habit before charging for it. Micky Irons, Mickai's founder, has said plainly that being free during beta will not necessarily stay true once the product matures. No date has been set for that to change, and no price has been decided, publicly or otherwise, as far as this article can state. If and when that changes, it should be announced as plainly as this decision is being explained now.

How an agent earns its place before release

Before any persona is published, its instructions and whatever source material it was built from go through a review: who actually owns the wording and knowledge involved, whether its system prompt could be talked out of its own limits, and a check for anything that should never appear in a public download, exposed credentials, hidden instructions, invented statistics presented as fact. Material with unclear ownership, or that cannot be checked properly, is held back rather than published anyway. That is also why the finished catalogue is smaller than the number of role ideas floating around Mickai's various projects: a role does not get released just because a name and a job title exist for it.

What you can actually do today

Right now, the honest next step is not downloading an agent, it is trying Trust Agent's free workbooks, which are genuinely live, and watching for the agents to move from final review to release. When they do, expect the same standard this article has tried to hold to: a plain description of what each one does, what it cannot do by default, and a published system prompt you can read for yourself rather than take on trust. Trust Agent sits alongside Mickai's wider work on sovereign, auditable AI systems; if that direction interests you more broadly, the Mickai beta is where it is currently being tested.

About this account: written by Mickarle Wagstaff-Irons - Micky Irons, full name Mickarle Sean Junior Wagstaff-Irons, Mickai's founder. Trust Agent's agent-review status was checked against the project's own build configuration on 29 September 2026. Unified and Trust Agent are part of the same Mickai publication family; this is not an independent review.