Paraline documentation

AI can already read documents, summarize conversations, research a market, draft a response, and recommend what to do next. The harder question is what happens after the answer appears.

Who checks it? Where is it saved? Which customer or project does it belong to? Can the AI take the next action, or should a person approve it? If something goes wrong, can the team see what happened and continue the work?

Paraline is being designed around those questions. It is a framework for bringing people, business data, software, and AI agents into the same operation. The aim is not simply to produce more AI output. It is to help a business complete useful work with clear rules, shared records, and a reliable history.

Why this documentation exists

AI terminology can make a practical subject feel unnecessarily complicated. Words such as agent, runtime, model routing, and MCP are useful, but they should not be the price of admission.

This documentation begins with the business problem behind each concept. It is written for owners, operators, consultants, and technical teams who want to understand:

• where AI can genuinely improve an operation;
• how an AI-assisted workflow differs from a chatbot;
• which decisions should remain deterministic or human;
• how data, permissions, approvals, and history make automation dependable;
• and how to introduce autonomy gradually, based on evidence.

The later sections include more technical detail for implementation teams. You do not need that detail to understand the operating model.

A simple example

Imagine a service company that receives requests by email, web form, and PDF. Today, an employee reads each request, identifies the customer, copies details into a spreadsheet, decides who should handle it, drafts a reply, and follows up until the issue is resolved.

A basic AI assistant could summarize the email or draft the reply. That saves some writing time, but the employee still has to move the result through the business.

A complete AI-assisted operation goes further:

• the request becomes a trackable case;
• the customer and relevant history are identified;
• important facts are extracted with links to their source;
• rules determine priority and ownership;
• AI prepares the parts that require interpretation;
• a person reviews uncertain or consequential decisions;
• approved actions update the right system;
• and the outcome remains visible to the team.

That surrounding structure is where Paraline fits.

The three ideas behind Paraline

Intelligence helps interpret the work

AI models are good at working with language and other unstructured material. They can understand a varied request, compare documents, extract information, draft content, and select from a set of tools.

Their answers are not perfectly predictable. That makes them valuable for judgment-heavy work, but it also means important outputs need evidence, validation, or review.

The operating framework keeps the work dependable

The framework holds the parts that should not disappear when a conversation ends: customers, cases, projects, permissions, approvals, workflow progress, and activity history.

In the technical sections, this durable operating layer is called the runtime. In plain language, it is the part that remembers what is true, decides what is allowed, and records what actually happened.

The application is what the team uses

People work through familiar interfaces: tables, forms, queues, dashboards, detail pages, and review screens. An application might look like a CRM, service desk, document-review workspace, or a specialized operations tool.

People and AI can use different interfaces while working with the same records and rules. An operator may click an approval button; an agent may request the same approved action through a tool.

What makes an AI workflow trustworthy

A useful workflow does more than return a polished answer. It should make the following questions easy to answer.

What started the work?

A message, upload, deadline, form, or request creates one identifiable responsibility.

What information was used?

The workflow retrieves relevant, permitted records and preserves the sources behind important claims.

What may the AI do?

The agent receives a narrow set of tools and permissions suited to its responsibility.

Where does a person decide?

Uncertain, sensitive, or high-impact actions pause for a clear review rather than hiding approval inside a conversation.

What counts as complete?

Completion is tied to a business result—such as an approved document or a delivered response—not to the moment an AI stops generating.

Can the result be explained?

The system retains the sources, proposals, decisions, actions, errors, and outcome needed to understand the work.

Where to begin

If you are a business owner or operator, start with What is Paraline?. Then read the core mental model, which explains the difference between AI intelligence and the business system around it.

To explore automation, continue to AI and automation and Workflow anatomy. The service operations example shows how the concepts come together in a familiar business setting.

If you are planning an implementation, the sections on data and imports, security and governance, and reliability and evaluation are especially important.

Technical readers can continue into How the operating layer works, Model Context Protocol, and Developer interfaces.

Vision and current availability

Paraline is in active development. Some pages describe the intended product architecture, not a feature that is already publicly available. The documentation uses four status terms:

• Available means the capability exists within the documented public experience.
• Prototype means it demonstrates an idea but is not a production guarantee.
• Planned means it is intended but not released as a stable interface.
• Concept explains the longer-term product model.

The capability status page explains the current boundary. It is the appropriate place to check before making an implementation or procurement decision.

The vision is ambitious, but the operating principle is simple: use AI where it adds judgment and speed, use software rules where consistency matters, and keep people in control of the decisions that carry real consequence.