Managed by Alian Software
AI

AI Glossary for Project Managers: 32 Terms in Plain English

An AI glossary for project managers: 32 terms from LLM and token to agent, MCP and RAG, explained in plain words with how each shows up in AlianHub.

AlianHub Team · 6 min read

This AI glossary for project managers explains 32 terms you will hear in AI conversations, in plain English, with a note on where each one shows up in AlianHub. You do not need to be technical to use AI well. You do need to know what "agent", "MCP" or "token" means when someone asks you to approve a budget.

Key takeaways

  • You do not need to code to manage AI. You need the vocabulary.
  • An LLM writes. An agent acts. The difference is what it is allowed to change.
  • MCP is how the AI you already use plugs into your tools.
  • Tokens are how AI is metered, which is why spend caps matter.
  • "Human in the loop" should be a setting you can point to, not a slogan.

The basics

1. AI (artificial intelligence). Software that does tasks that usually need human judgement, like writing, sorting or summarising.

2. LLM (large language model). The model behind tools like Claude and ChatGPT. It predicts text, which lets it answer, draft and plan. In AlianHub, the model is the only external part.

3. Model. A specific LLM. Different models trade speed, cost and quality. AlianHub lets you choose the model per question in the Ask box, or pin one per agent.

4. Provider. The company that runs the model. AlianHub's AI Assist runs on your own LLM API key, with multiple providers supported.

5. API key. A secret that lets software call a provider's model and bills you for it. Treat it like a password.

6. BYOK (bring your own key). Using your own API key rather than a vendor's bundled AI. You see and control the cost.

7. Prompt. The instruction you give a model. "Plan a two-week sprint for the pricing page" is a prompt.

8. Token. The unit models count text in, roughly a piece of a word. Providers charge per token, so long inputs cost more.

9. Context window. How much text a model can consider at once. Bigger windows read more, but cost more.

10. Hallucination. When a model states something false with confidence. The reason drafts need review.

Working with AI in projects

11. Assistant. An AI that answers and drafts when you ask. AlianHub's Ask box is this: plain words, scoped to projects you can open.

12. Agent. An AI that takes actions toward a job, like commenting or creating subtasks. In AlianHub, an agent is a workspace member with an owner, a run budget and a scope, marked with an AGENT tag.

13. Skill. A playbook an agent follows, such as a digest or a brief parser. AlianHub skills are versioned and can be dry-run, edited or retired.

14. Dry run. Previewing what a skill would do without writing anything.

15. Allowed actions. The list of things an agent may do. Anything outside it cannot run.

16. Action registry. The full set of actions that exist for agents. In AlianHub, destructive actions like deleting a project are absent from it, not merely switched off.

17. Autonomy level. How much an agent may do without asking. AlianHub uses L0 to L3.

LevelMeaning in AlianHub
L0Answers and suggests. Changes wait in the AI Inbox
L1Proposes every change. Nothing is written until approved
L2Runs low-risk, reversible changes. The rest waits
L3Like L2, and may also run on a schedule

18. Human in the loop. A person approves an AI's action before it takes effect.

19. Approval queue. Where proposals wait. In AlianHub, it is the AI Inbox.

20. Audit log. A record of every action. AlianHub logs whether an agent acted, proposed or was refused.

21. Undo. Reversing a change. Changes an agent makes on its own in AlianHub come with Undo.

22. Spend cap. A limit on AI cost per period. AlianHub agents have a monthly spend cap and a daily action limit.

23. Routing policy. A workspace rule for which model handles which work.

24. Router. In AlianHub, it proposes an agent for each selected task and refuses those that need a person.

25. Team pack. A set of role agents for one kind of team, turned on in one click and starting paused.

26. AI memory. Facts and preferences the AI keeps about you. AlianHub's is private to you, up to 50 items.

Connecting AI to tools

27. MCP (Model Context Protocol). An open way for AI apps to use tools. AlianHub ships an MCP server, so Claude, ChatGPT or Claude Code can work inside your workspace.

28. Connector. The entry you add in your AI app to reach a tool. For AlianHub, you paste an address as a custom connector.

29. Acting as you. A connected AI uses your identity and permissions, so it can open only what you can.

30. RAG (retrieval-augmented generation). Giving a model relevant documents at the moment it answers, so it works from your facts. Ask answering from the projects you can open is the same idea in practice.

31. Embedding. A numeric version of text used to find similar content. Often part of RAG.

32. Self-hosted. Software you run on your own servers. AlianHub is self-hosted and open source under AGPL-3.0.

Tip: share this glossary with your team before your first AI rollout meeting. Half of AI disagreements are two people using one word differently.

How to do this in AlianHub

Try the terms for real in the live demo:

  1. Ask (term 11). Open the Ask box, keep "Everything you can open" and ask "Find my urgent work".
  2. Skill and dry run (13, 14). Open the Skill library, pick Intake, and run a Dry run on a task brief.
  3. Autonomy level (17). Open the Agents tab and look at each agent's level.
  4. AI Inbox (19). Approve or decline a proposal, then find it in the audit log (20).
  5. MCP (27, 28). Open AI, then Connect your AI, and copy the address for your app.

A business scenario

A new project manager joins a manufacturing team that already uses agents. In their first week, they hear "L2", "spend cap" and "MCP" in one meeting.

With this glossary and a walk through the Agents tab, they can read an agent's settings, approve its proposals with confidence, and connect their own Claude account to ask about supplier tasks. Onboarding takes less of the team lead's time, and the new manager contributes sooner.

The best AI skill for a project manager is knowing which question to ask about an agent.

See all of these in context on the AI page and the AI controls page.

FAQ

What is the difference between an LLM and an agent?

An LLM produces text. An agent uses a model to take actions, like commenting or creating subtasks, within the limits it is given.

What does MCP mean for project managers?

It means the AI app you already use can work inside your project tool. In AlianHub, it acts as you and sees only what you can open. See Connect your AI.

Why do tokens matter for my budget?

Providers bill by tokens, so heavy AI use costs more. Spend caps and action limits keep that predictable.

Is RAG built into AlianHub?

We do not describe it that way. Ask answers from the projects you can open, which is the practical idea behind RAG.

Bookmark this AI glossary for project managers, try the terms in the live demo, and share it with your team before the next AI meeting.

  • #AI glossary for project managers
  • #AI terms
  • #MCP
Try AlianHub on your own serverFree and open source under AGPL-3.0. No per-seat pricing.
Explore the Live Demo
Wait! Before you go…

Try AlianHub freefor 7 days

Get your own private AlianHub, hosted and managed by us. Start the free trial on the Starter plan and bring your whole team.

  • Every feature included, nothing locked
  • Unlimited users and guests
  • Hosted, secured and updated for you
Start my free trial

7-day free trial on the Starter plan

View Brochure