All use cases
Agentic

Knowledge Agent

An agent answers employee and customer questions from your docs, wikis, and past tickets, citing sources and flagging gaps in the documentation.

ConfluenceZendeskSlack

The problem

The answers exist somewhere: in Confluence, in old tickets, in a wiki page nobody remembers. People ask the same questions in Slack anyway, because searching several systems is slower than asking a colleague. Support staff and internal experts keep re-answering questions that were already solved.

How it runs on Neblex

Neblex indexes your Confluence pages and past Zendesk tickets into your own Postgres/pgvector database, with embeddings that can run on self-hosted models. When someone asks a question, the agent runs semantic search over that index, composes an answer, and cites the sources it used, so readers can verify every claim against the original page or ticket.

The agent meets people where they ask: Slack, Microsoft Teams, embeddable web chat, and voice conversations. Durable memory carries context across conversations, so follow-up questions do not start from zero.

When coverage is thin, the agent flags the gap and routes it to documentation owners by opening a ticket or posting to a Slack channel through connectors. Every answer and its citations are logged to the tamper-evident hash-chained audit trail.

Step by step

1

Index your content

Confluence pages and past Zendesk tickets are indexed into your own Postgres/pgvector database.

2

Ask in any channel

People ask in Slack, Teams, embeddable web chat, or by voice.

3

Agent searches and answers

Semantic search finds the relevant passages and the answer cites its sources.

4

Gaps get flagged

Questions without a grounded answer route to documentation owners through a ticket or a Slack post.

5

Memory carries context

Durable memory keeps context across conversations so follow-ups build on earlier answers.

Platform capabilities used

  • RAG over your own Postgres/pgvector
  • Source-cited answers
  • Slack, Teams, web chat, and voice channels
  • Durable agent memory
  • Self-hosted embedding models
  • Hash-chained audit trail

Common questions

Where does our content live once it is indexed?

In your own Postgres database with pgvector, not in a third-party vector store. Embeddings can run on self-hosted models, and On-Prem Workers can run the pipeline inside your network over an outbound-only connection.

How do we know an answer is right?

Answers cite their sources, so anyone can check the underlying page or ticket. Like any LLM output, answers should still be verified for high-stakes decisions, which is why citations are attached and every response is logged to the audit trail.

Can we control what the agent is allowed to do?

Yes. You control what gets indexed, role-based access governs who can change the agent, and any action beyond answering, such as opening a ticket, can require human approval. Every response is logged to a tamper-evident hash-chained audit trail.

Want this running on your stack?

Neblex Integration Fabric is in beta: full platform, free while in beta. Bring this workflow and we will map it to your systems.