What an AI agent is, and what it is made of
The vocabulary without the hand-waving — what an agent actually is, what makes one fit to face your customers, and what each piece in your console does.
Start here
What the thing actually is — and why "we already have GPT" is not the same sentence.
What is an AI agent?
Software that holds a conversation, looks things up in your documents, and takes actions on your systems — deciding for itself which to do next and following a task through, step by step, rather than only replying to one message at a time.
ReadAI agent vs LLM: what is the difference?
The model is the engine that turns text into text; the agent is everything built around it — retrieval, tools, memory, boundaries and channels. Models are replaceable parts inside an agent.
ReadWhat is a business AI agent?
An agent that works for one specific business — answering only from that company's own documents, staying inside a scope its owner defines, remembering each customer across visits, and accountable for what it says. Not a general chatbot, but a member of that business's team.
ReadAgent harness: the layer that turns a model into a dependable agent
The engineering layer wrapped around a language model — retrieval, tools, skills, memory and guardrails — that turns a general model into a specific, dependable agent. The model supplies fluency; the harness supplies grounding, boundaries and the ability to act, and stays the same when the model is swapped.
Readagent4.io as headless software
Headless software exposes everything over an API, MCP and skills so that an AI agent — not only a human clicking a UI — can operate it. You describe the outcome; your agent, handed exactly the right skills, drives the features precisely.
ReadAgent-to-Agent (A2A): when your customers' AI assistants do business with your agent
A machine-to-machine channel where a customer's personal AI assistant talks directly to a business's agent — consulting it and submitting a structured request — instead of a person typing into a chat box. agent4.io handles the pre-processing half and hands back a summary, while a human confirms anything irreversible.
ReadWhat it is made of
The five things you configure in the console, and which job each one does.
What is an agent's soul (system prompt)?
The standing description of who an agent is — its character, tone and non-negotiables — held separately from the job it happens to be doing.
ReadWhat is a knowledge base for an AI agent?
The documents an agent may answer from. agent4.io keeps two kinds — your company's vetted material, and each customer's own files — retrieved together but never confused for one another.
ReadWhat is an agent Skill?
A named capability pack — a short description of when to use it, fuller instructions, and the tools it unlocks — that the agent loads only when it becomes relevant.
ReadWhat is MCP (Model Context Protocol)?
The Model Context Protocol — an open standard for connecting an agent to external tools and data, so a system exposes its capabilities once instead of once per AI product.
ReadKnowledge base, tools or Skill — which does what
Three different jobs that get confused for one — the knowledge base holds facts, an MCP tool performs a calculation you cannot let a model improvise, and a Skill is the procedure that decides which to reach for.
ReadFine-tuning vs RAG: when you actually need to fine-tune
Continuing a model's training on examples of the behaviour you want, so that behaviour becomes its default instead of something you ask for on every request. It changes how a model works, not what it knows.
ReadHow it behaves
What the agent does on its own — before anyone types, and after everyone leaves.
What is a page playbook (page-aware AI chat)?
A per-page briefing for the agent — background it should already know, an opening line, and a few suggested questions — chosen automatically from the URL the visitor opened the chat on.
ReadWhat are scheduled tasks for an AI agent?
Work an agent commits to for a specific future moment — a follow-up it decides to send, or a recurring reminder — executed on time by a scheduler rather than waiting for someone to speak first.
ReadWhat are ask forms (forms inside the chat)?
A small form the agent composes mid-conversation — choices, a number, a date — rendered as something to tap rather than a paragraph of questions to answer by typing.
ReadWhat is a space (agent workspace)?
The isolated container holding one customer's conversations, documents and memory — enforced at the database level, so one customer's material cannot surface in another's conversation.
ReadPrompt injection: when the visitor tries to rewrite your agent
An attack in which text supplied by a user — or hidden inside a document, web page or email the agent reads — is treated by the model as a new instruction rather than as content, causing the agent to abandon its rules, change its identity, or reveal its system prompt. It cannot be fixed by writing a stricter prompt, because the attack arrives through the same channel as legitimate input; the defences that work sit before the model and after it.
ReadGetting it running
The part that stalls most pilots, and why it does not have to.
How to build a knowledge base for an AI agent
Most of what an agent needs to know is already written down — on your website, and in the PDFs your team already sends customers. Importing from a URL covers the public half; uploading covers the rest.
ReadStructured index: how an AI agent answers "how many" about your documents
A structured index is a small table an agent builds from the fields your documents already carry — title, price, level, link, image — alongside the vector search it uses for meaning. Vector search finds passages that read like the question; it cannot count, filter by number, or group. The structured index answers those, and lets the agent quote a link or an identifier exactly instead of reconstructing it from a fragment.
ReadStorylines: stateful, graph-orchestrated agents that deliver a process
A storyline is a directed graph an agent follows with each end user — every node is a step (its own task, knowledge and tools), exits carry conditions, and each person's progress, profile and notes are saved and resumed across sessions and channels. It turns an agent from an assistant that does point-tasks into one that can deliver a multi-step service itself.
ReadDynamic Planner: AI that turns a complex request into a step-by-step plan
The Dynamic Planner watches a conversation and, when a user is pursuing a genuinely multi-step task inside the agent's domain and is visibly unsure how to proceed, offers to expand that task into a temporary storyline — a checklist-driven, step-by-step plan the agent then executes one focused step at a time, with progress the user can see. Planning happens once, under validation; execution is deterministic.
ReadLong-form writing: an agent that plans, researches and writes a real document
Long-form writing is an agent producing a complete, multi-section document — a filing, a market-entry report, a due-diligence memo — by planning the outline first, asking for anything it needs before it starts, researching each section against your material and public sources, and writing section by section on a job that survives a restart. You edit it in a canvas beside the chat, rewrite any passage by selecting it, and every version is kept.
ReadStill not sure which piece you need?
Tell us what your business does and we will say which of these you would actually use — and which you can ignore.
Read the docsTalk to us