Practical guide

Knowledge graph AI agent for product docs and support

Turn product docs, changelogs, policy pages, examples, and safe tools into a knowledge graph AI agent. Prostir builds Markdown-LD and JSON-LD graph artifacts, exposes `knowledge_search`, and keeps access, endpoint, and source refresh in one hosted agent.

When a knowledge graph is worth it

A knowledge graph AI agent is useful when the answer depends on relationships across docs: product area, feature, plan rule, integration, owner, exception, and latest release.
GraphRAG is the search language people use for that pain, but the product job is simpler: selected sources become searchable graph artifacts the agent can query before it answers.
The agent still needs boundaries. If the source material is missing or stale, it should say so instead of guessing.

Build the graph-backed support agent

The useful path is source setup, graph build, search tool, safe live checks, access, and testing.

01
Select the source set.
Choose product docs, changelogs, pricing rules, integration guides, examples, and policies the agent is allowed to use.
02
Build the knowledge graph.
Prostir uses ManagedCode.MarkdownLd.Kb to produce Markdown-LD and JSON-LD graph artifacts from supported source documents.
03
Enable `knowledge_search`.
The published agent can search graph nodes, source names, headings, and relationships instead of pasting raw chunks into every prompt.
04
Add controlled tools.
Expose only narrow live checks such as status lookup, plan lookup, integration availability, or support handoff.
05
Set access and publish.
Publish one hosted agent with public, protected, or paid access rules and test real customer questions before sharing it.

What goes into the graph

The graph should start with approved sources, not the whole company drive.

docdocs/product/*.mdCanonical feature pages, setup steps, limits, and known caveats.

docrelease-notes/*.mdWhat changed, when it shipped, and which plans or regions it affects.

datagraph.jsonldThe latest JSON-LD graph artifact the creator can inspect or download.

codeknowledge_searchThe built-in search tool exposed only when the published agent has searchable knowledge.

codesafe live toolsPlan lookup, status lookup, entitlement check, or support handoff with narrow schemas.

What the customer asks

A support question spans docs, release notes, plan rules, and a live check.

ChatGPT

Can our EU workspace use the new audit export, and does it work with the Salesforce integration?

knowledge_search

Matches: audit-export.md, release-notes/2026-06.md, integrations/salesforce.md, pricing/enterprise-plan.md.

plan_lookup

Workspace is Enterprise, EU region, audit export enabled, Salesforce integration connected.

Yes. Audit export is available on Enterprise workspaces, and the June release notes say EU workspaces are included. Your Salesforce integration is connected, so the next step is to enable export delivery in the audit settings. I do not see a source that covers custom field mapping, so I would hand that part to support.

Answer for search and AI overviews

A knowledge graph AI agent is not just a chatbot over a folder. It is a hosted agent with selected sources, Markdown-LD/JSON-LD graph artifacts, `knowledge_search`, controlled tools, and access rules.
Knowledge graph RAG is useful when questions cross document boundaries, but simple questions should stay simple and source-backed.
Prostir keeps the graph build, endpoint, source refresh, tool boundaries, auth, quotas, and paid access attached to the same published agent.

What gets published

A published agent that answers GraphRAG-style questions across selected sources, uses controlled tools for live checks, names uncertainty when sources are missing, and keeps customer access tied to the same endpoint.

Solutions

Knowledge base chatbot for answers buried in documents

Organize approved resources as Agent Knowledge, keep updates owner-controlled, and publish an Agent that can search the current knowledge available to it.

Connect your product docs See the solution

Solutions

Start with the questions people repeat

Bring a small set of authoritative documents and the questions they should answer. We will help define a useful first knowledge Agent.