How Experio Works

​

The short version

Experio connects to a client’s document libraries and business systems and reads what is there. It turns that content into a knowledge graph: one connected record of the things the business cares about, such as clients, projects, people, skills, contracts and obligations, and how they relate. Users ask questions in chat, or run agents that produce documents. Experio answers from the graph and cites the files and records behind each answer. Everything in the middle column is driven by configuration that the implementation team agrees with the client: the ontology, taxonomies, artifact types, data mappings, matching strategies and enrichment rules. The software does the reading. The configuration tells it what to look for and how to connect it.

​

The components

You don’t need to operate these services, but it helps to know they exist when you read job status or talk to Experio support.

​

Building the graph

There are two ways in, and they work differently. How Data Gets to the Graph covers both in detail.

• Documents (proposals, contracts, resumes, reports): an AI model reads each file and extracts only what the artifact type asks for. It matches what it finds against existing records, using similarity, AI review and human review.
• Tables (exports from HR, CRM and finance systems, or API responses): a data mapping tells Experio exactly which column becomes which entity, attribute or relationship. Matching is by exact key.

Both land in the same graph, on the same entity types. The goal of an implementation is that “Lakeshore Health” from the CRM, from a folder name and from the text of a contract all end up as one Client node, with every source recorded in its lineage.

​

Using the graph

​

Chat assistants

When a user asks a question, the assistant:

1. Routes the request. It decides whether this is a question about internal data, a request that also needs external tools such as email or CRM, a follow-up to the previous answer, or a job for an agent flow.
2. Resolves names in the question to graph records. “Lakeshore” becomes the Client node Lakeshore Health. This uses exact matches and the similarity (vector) indexes set per entity type on the ontology’s Indexes tab.
3. Plans a query. It works out which graph paths answer the question, writes a graph query from the ontology and taxonomies, and checks the query before running it.
4. Gathers context from the matched records and their neighbours, and optionally from document text.
5. Writes the answer with citations. Users can open each citation and trace its lineage back to the file or row it came from.

Citations respect access control. A user who can’t see a contract won’t see it cited.

​

Agent flows

An agent flow is a multi-step AI workflow built on a visual canvas. A typical flow collects inputs, pulls data from the graph, has a model rank or draft something, pauses for a human to approve, and produces a Word, PowerPoint or Excel file. Users launch flows from chat. The assistant can also start one when a request matches the flow’s description.

​

Other ways in

• Document templates: branded PowerPoint and Word output from chat, through /generate, the wand button, or by asking for a document type.
• MCP server: users can call Experio from Claude Desktop, Cursor and other MCP clients with a personal API key.
• Integrations: per-user Google Workspace and Slack connections, so an assistant can combine graph answers with the user’s email, calendar or channels.

​

Where the configuration lives

The admin panel (/admin) groups its pages by stage of the work. Most implementation effort is in the first three sections. Every admin page has an Admin Copilot (⌘/Ctrl + J). It answers questions from this guide and the Admin Guide, and on some pages it can draft changes for you to review.

​

Next