The New ARMES Concierge: Your AI Workspace Architect

There's a gap between signing up for an AI platform and actually getting good results from it. Not a technology gap — a setup gap.

Most people land, open a chat, and start prompting. It works. But they're leaving enormous value on the table: the custom instructions that shape how AI responds to them, the knowledge base that gives AI memory of their world, the project organization that keeps context clean across different workstreams. The stuff that turns a generic chatbot into a personal AI ecosystem.

Setting all of that up well is hard. It's context engineering — the kind of work that prompt engineers and AI consultants charge for. Until now, the ARMES Concierge could advise you on how to do it. Today, it can do it with you.

What changed

The Concierge was always the setup specialist in ARMES — the agent that knows every feature, every surface, every configuration option. It could explain what to do. It could recommend what to set up. But it couldn't build anything. You'd get advice, then go do it yourself.

That limitation is gone. The new Concierge can create note folders, draft and save notes, audit your custom instructions across every agent and project, review your knowledge base structure, and walk you through the entire setup — all inside the conversation, all in your voice, all with your approval before anything is saved.

The role shifted from consultant to architect. You talk. The Concierge listens, asks the right questions, and builds.

Your knowledge base, built in conversation

The most impactful thing the Concierge does now is construct your foundational knowledge base. Here's what that looks like.

You open a conversation with the Concierge. It asks you a few questions — not a form, just a conversation. Who are you? What do you do? What are you working on? How do you prefer AI to respond?

Then it gets to work. It creates your note folders — organized by your domains, projects, or however your work naturally breaks apart. It drafts your foundational notes: an "About Me" note with your professional context and goals, an "AI Preferences" note with your communication preferences and expertise level, project-specific notes for whatever you're building. You review each draft, refine it, and the Concierge saves it.

Those notes become the raw material that every agent in your ecosystem can search and reference. Your ARMES GPT agent pulls your "About Me" when you ask it to draft a bio. Your Custom Agent reads your "Acme Corp - Client Brief" when you're preparing for a meeting. The knowledge lives in one place and works everywhere.

For users who already have reference documents — brand guides, contracts, SOPs, spreadsheets — the Concierge guides you on where to upload them as persistent files so your agents can search and read them in any conversation.

Instruction intelligence

Custom instructions are the most powerful personalization tool in ARMES — and the most commonly misunderstood. Users tend to fall into one of two traps: writing nothing and missing the value, or writing too much and actively degrading their AI's performance.

That second one is counterintuitive, but the research is clear. Models process every token of instruction against every token of your message. More instructions means more competition for the model's attention. Verbose, over-specified instructions cause models to shift from deep reasoning to surface-level compliance. The model is so busy following rules that it stops thinking.

The Concierge knows this. It has a unique tool that no other ARMES agent has: the ability to audit all of your custom instructions — global and per-agent — across every project. It sees the full picture.

When you ask the Concierge to review your setup, it evaluates your instructions against a set of principles:

• Is this knowledge or behavior? Paragraphs of domain knowledge embedded in instructions belong in notes. Instructions should shape how an agent behaves, not what it knows.
• Is this duplicated? The same instruction copied across five agents should be a single global instruction.
• Is this bloated? If a per-agent instruction exceeds a few hundred characters, something probably belongs in a note with a conditional reference.
• Is context scattered? The same information maintained in multiple instruction fields should be centralized in one note that multiple agents reference — update once, every agent benefits.

The Concierge drafts lean replacements, shows you before-and-after comparisons with estimated token savings, and offers to create any notes that should replace embedded knowledge. You copy the new instructions into Settings; it verifies the change.

The layered strategy

One of the most valuable things the Concierge teaches is what we call the layered strategy — using the right tool for the right kind of context:

Prompt templates with pinned notes are the most efficient option. A "Weekly Review" template with your "Active Projects" and "Goals" notes pinned to it delivers exactly the right context when you use it, and costs zero tokens when you don't. One click, perfect context, no waste. This is the single most underused feature on the platform.

Lean conditional instructions handle the unpredictable. A line in your per-agent instructions — "client onboarding → read note 'Client Onboarding'" — tells the agent to pull that note only when the topic comes up. A few keywords, not a paragraph. Maximum five conditionals per agent.

Global instructions are the baseline. Three to five lines that define who you are and how you prefer AI to respond, applied to every agent in the active project. If it's agent-specific, it doesn't belong here.

Notes and files are the knowledge store. This is where the actual information lives — your preferences, project briefs, domain knowledge, reference documents. Instructions point to notes. Notes hold the knowledge. Update a note once; every agent that references it benefits automatically.

The Concierge doesn't just explain this hierarchy. It applies it to your specific setup and shows you where your context is in the wrong layer.

Custom agents, configured with help

Earlier this year we introduced the ARMES Custom Agent — a blank-canvas agent that you define entirely through custom instructions. It's the only agent in ARMES that can be installed multiple times, each instance with its own identity, personality, and context.

Users have built some genuinely creative things with it. One consultant created a dedicated Custom Agent for each of their clients — a per-client relationship manager with its own instructions and note references. A writer built separate agents for different book projects. A small agency uses instances for each department.

But writing the instructions that bring a Custom Agent to life is exactly the kind of context engineering most people aren't sure how to approach. What should go in the instructions? What shouldn't? How do you reference notes? How specific should the persona be?

This is where the Concierge excels. Tell it what role you need — a client advisor, a technical reviewer, a writing partner, a domain specialist — and it interviews you, drafts the instructions, creates the supporting notes, and presents everything in a code block ready to paste. It understands that Custom Agent instructions should define who the agent is (role, persona, tone, domain focus), not what it can do (the platform provides the capabilities automatically). It keeps instructions lean and puts domain knowledge in notes where it can be updated without touching the agent's configuration.

If you need an agent and nothing in the Agent Library fits, the Concierge's first move is to suggest a Custom Agent and offer to build it with you.

Phased, not overwhelming

A full onboarding — knowledge base, instructions, agent recommendations, templates — involves a lot of decisions and a lot of content. The old approach tried to do it all in one marathon session. The new Concierge doesn't.

Large workflows are broken into phases of two to three deliverables. After each phase, the Concierge summarizes what was completed, states what's next, and asks: "Ready to continue, or want to pick this up later?"

If you step away, the Concierge offers to save a progress note — a "Setup Progress" note listing what's done and what remains. When you come back — even in a new conversation — it picks up where you left off. It re-orients by checking your current setup rather than re-asking questions the platform already has answers to.

This isn't just a technical safeguard. It's better UX. Nobody wants a setup marathon. You want to make progress, feel good about it, and come back when you're ready for more.

Who the Concierge is (and isn't)

A word on scope, because this matters.

The Concierge is a specialist. Its job is to build, configure, and optimize your ARMES workspace. It is not a general-purpose assistant. It won't write your marketing copy, debug your code, or conduct your research — it'll point you to the specialist agent that will.

What it will do is make every one of those specialist agents better. The instructions it helps you write, the notes it helps you create, the templates it helps you design, the organizational structure it helps you build — all of that is the foundation that every other agent conversation draws from. A well-configured workspace is the difference between an AI that feels generic and one that feels like it was built for you. Because it was. By you and the Concierge, together.

Getting started

The Concierge is available on every plan, including Free. Switch to it by typing #ARMES Concierge in any chat, or find it in your agents list.

A few good opening moves:

• New to ARMES? Just say hello. The Concierge will start a conversation, learn about you, and begin building your workspace.
• Been here a while? Ask it to audit your setup. "Review my instructions" or "optimize my workspace" will trigger a full ecosystem review.
• Want a custom agent? Tell the Concierge what role you need. It'll draft the instructions and supporting notes.
• On the Command page? Even better. You can have the Concierge chat, your notes, and your settings open side by side — ideal for setup work.

Your AI is only as good as the context it has. The Concierge makes sure it has the right context, in the right place, in the right amount.

Joseph Founder, ARMES