Neuronz.ai

English

Français

English

Français

The knowledge graph ​

Facts have evidence, incidents have causes, fixes belong to the work that shipped them. The knowledge graph records those relationships as named, one-way links between your records, so a single fact can lead your agent to everything connected to it.

How it works ​

Links between any two records. A fact, a past action, a knowledge doc, a topic, or a person or group from your social map can be linked with a relationship that says how they relate:

  • evidenced by: the source that backs a claim,
  • caused by / fixed by: the cause of an incident, the change that resolved it,
  • derived from: a fact taken from a larger doc,
  • member / relates to / supersedes: grouping, association, and history.

Links stay within a profile, like everything else.

Recall follows them. When your agent reads the records that match its prompt, recall also brings back the records directly linked to them, one step out. This happens only when your agent reads records, not on every prompt.

Some links are made for you. A fact saved from a knowledge doc is linked to that doc, for example. Your agent can add others as it works.

Related topics. Two topics linked with "relates to" show up in each other's detail panel as related topics, so connected threads point to each other.

Who belongs to what. The people and groups in your social map can be linked with "member": Sam is one of your clients, Dana is on the design team. Voice uses that link: a writing style you set for clients also applies to a message to Sam, without setting one up for each person. Say it once, when you first describe someone.

Using it ​

Your agent manages links over MCP: link and unlink to create or remove one, and expand to walk outward from any record.

Mostly you notice it in recall: ask about a retry policy, and when your agent reads the matching fact it also gets the incident that caused it and the runbook that documents it, because they were linked once.