Skip to main content

​
Overview

Librarian is the AI assistant built into Neuradex. It’s not just a chatbot—it answers questions while referencing all knowledge in your project and automatically records new knowledge discovered through conversations.

​
Features

​
Knowledge-Powered Conversations

When Librarian receives a question, it automatically searches knowledge within the project.
  • Knowledge: Prioritizes confirmed facts
  • SourceContent: Extracts relevant information from documents
  • Episode: Can also reference past conversations and work history
It generates well-grounded answers based on search results.

​
Automatic Knowledge Recording

Facts discovered during conversations are automatically recorded as Knowledge by Librarian. 
User: "We've decided to use JWT for authentication."
Librarian: "Understood. I've recorded the authentication method as JWT."
           → Knowledge "Authentication method is JWT" is auto-created

​
Conversation History

All conversations are automatically saved as Episodes. You can search for “what we discussed” later.

​
Context Management

Even in long conversations, proprietary technology enables efficient memory utilization. Relevant information is properly retained while unnecessary information is automatically organized.

​
Autonomous Agent Behavior

Librarian doesn’t just answer questions—it autonomously selects and calls tools in iterative loops. Under the hood, a proprietary agent architecture drives the Librarian, where the LLM autonomously selects and executes tools in a loop. For a single question, it automatically performs multi-step processing: search → analyze → additional search → respond.

Semantic Knowledge Search

Understands query intent and explores related knowledge through multi-hop traversal

Auto-Extract & Update Facts

Detects new facts from conversations and suggests linking to existing knowledge

Change History Tracking

Records knowledge changes as episodes, tracking “who changed what, when”

Auto-Discover Relationships

Automatically detects references, extends, contradicts, and supersedes relationships between knowledge

​
Approval Flow

Before adding, updating, or deleting knowledge, Librarian always shows a proposal and asks for user approval. 
Librarian: "Should I record the authentication method as JWT?"
           [Approve] [Edit] [Reject]
It never changes knowledge without permission. Autonomous yet controllable by design.

​
Use Cases

​
Understanding Project Status

Cross-searches multiple knowledge items and Q&A history to summarize the latest status, such as “What was decided last week?” or “What tasks are incomplete?” Finds relevant documents and extracts the needed information — for example, API authentication details or deployment procedures.

​
Recording Knowledge

Extracts facts from conversations (e.g., “We’ve decided to use JWT for authentication”) and registers them as knowledge after approval.

​
Librarian Workflow

​
Difference from Chatbots

LibrarianChatbot
PurposeUtilize project-wide knowledgeCustomer support for specific use cases
AccessDashboard, API, MCPWebsite-embedded Widget
Conversation RecordingSaved as EpisodesSaved as Episodes
Auto Knowledge CreationYes (extracts facts from conversations)No
UsersProject membersEnd users
Librarian can also be used by AI agents like Claude through the MCP server. It serves as long-term memory for AI agents.

​
Sessions

Conversations with Librarian are managed in “sessions.” Context is maintained within the same session, and you can start a new session for each topic. Access programmatically via the Librarian API.

​
Next Steps

Data Types

Understand Knowledge & Episode

Librarian API

How to access via SDK