# Architecture & capabilities ## Overview The Knowledge Manager is built on five agentic skills that work together to deliver contextual knowledge retrieval and continuous knowledge improvement. This article covers the core skills, how the system is deployed and accessed, integration points, and how human oversight is maintained throughout. ## Agentic Skills #### Conversational Query Interpretation Understands user intent and context from natural language queries, identifying intent, urgency, asset type, and operational risk without relying on keyword matching. This skill forms a working hypothesis before any retrieval begins, enriching the query with asset and service history metadata, distinguishing troubleshooting requests from procedural queries, and routing to the most appropriate knowledge sources. If the query lacks sufficient context for a high-confidence response, it asks a targeted follow-up question rather than proceeding with incomplete information. #### Multi-Source Knowledge Search Searches simultaneously across multiple structured and unstructured systems to gather relevant information, manuals, engineering documentation, bulletins, historical work orders, case resolutions, similar asset failures, and prior repair outcomes. Unlike basic search, this skill cross-checks content consistency across sources, weighs recency and outcome history, and assembles a unified operational view rather than returning individual documents. #### Context-Aware Retrieval Fetches knowledge tailored to the specific asset, configuration, and operational scenario. Using asset ID, configuration variant, service history, and job context, this skill delivers configuration-specific troubleshooting steps and suppresses non-applicable or outdated instructions. The result is guidance that applies to the exact situation the user is facing, not generic documentation that requires interpretation. #### Relevance Ranking Evaluates and prioritizes validated, configuration-aligned guidance while suppressing outdated or low-confidence content. This skill ranks results based on source validation, usage signals, and resolution effectiveness across historical cases, returning guidance with clear confidence indicators so users understand how reliable each piece of guidance is. High-confidence, validated steps are surfaced first; unsafe or generic content is filtered out. #### Knowledge Gap Detection Identifies repeated queries, workarounds, and unresolved patterns to surface missing or weak documentation for continuous improvement. This skill monitors confidence levels and outcome signals from every interaction, detects recurring questions that the knowledge base cannot answer well, and automatically flags gaps for review. Where gaps are identified, it supports structured content creation, generating draft knowledge articles from field feedback and routing them through a human approval process before they are published. ## Deployment and Access The Knowledge Manager is a user-invoked conversational agent. It responds to questions raised directly by engineers, agents, or planners, it does not run as a background processing worker. **Primary interface:** Microsoft Teams, accessible on both desktop and mobile. Teams provides the conversational interface through which users submit questions, receive guidance, follow up, and rate responses. **Additional interfaces:** * API access for integration into existing customer chatbots or service platforms * Embedded chat widget deployable with a single line of code into any web interface * Deep integration into existing bots such as Salesforce Einstein, where the Knowledge Manager acts as a callable skill that receives context from the host bot and returns structured, governed responses This interface-agnostic design means the Knowledge Manager can enhance existing conversations and service platforms without requiring users to switch tools. ## Connected Knowledge Sources The Knowledge Manager connects to a wide range of knowledge sources across an enterprise: * **Manuals and technical documentation:** equipment manuals, engineering bulletins, safety notices * **Service management systems:** ServiceNow, Zendesk, Freshdesk, Salesforce Service Cloud * **Knowledge platforms:** Confluence, Notion, Guru, SharePoint, Google Drive * **ERP and field service systems:** IFS Cloud (FSM, PSO, service requests), Dynamics * **CRM systems:** Salesforce, HubSpot, Kustomer, Gainsight * **Collaboration and document tools:** Microsoft Teams, Google Drive, Poka * **Historical service records and case notes:** resolved tickets, prior repair outcomes, field job notes Sources are configured based on what is relevant and authorized for each deployment. The system can operate across any combination of structured and unstructured sources simultaneously. ## Integration Points **Collaboration Platforms:** Microsoft Teams and Slack serve as the primary conversational interfaces for field engineers and support teams. **Enterprise Applications:** IFS Cloud, Salesforce, and other enterprise platforms provide asset data, service history, and operational context that enriches retrieval. **Support Systems:** ServiceNow, Zendesk, Freshdesk, and similar platforms provide ticket history and case resolution data. **Knowledge Platforms:** SharePoint, Confluence, Notion, Guru, and Google Drive provide document-based knowledge content. **Email:** Outlook and Gmail support communication workflows where needed. **Documents:** IFS Loops and Google Drive support document retrieval and knowledge base management. ### Human-in-the-Loop Design The Knowledge Manager is designed with human oversight built in at the points where it matters most, particularly in high-compliance and safety-sensitive environments. Knowledge retrieval proceeds automatically when: * The query contains sufficient context for a high-confidence response * Matching knowledge exists across connected sources * The guidance is validated and configuration-appropriate Human input is required when: * The query lacks sufficient context and the system asks a clarifying question * A knowledge gap is detected and draft content is generated for review and approval * A response is rated negatively and the interaction is flagged for review * Knowledge from a resolved interaction is converted to a draft article requiring approval before publication Role-based visibility ensures users receive guidance appropriate to their role and access level. Escalation paths are maintained for cases that require senior review. Full auditability means every query, response, confidence rating, and outcome signal is tracked, supporting compliance requirements and continuous quality improvement. ### Feedback and Continuous Improvement Every response can be rated with a thumbs up or thumbs down directly in the interface. This feedback, combined with follow-up question patterns, resolution confirmation signals, and recurring gap detection, feeds continuously into improving both response quality and knowledge content. The feedback loop operates at two levels: improving how the system retrieves and ranks existing knowledge, and identifying what new or updated knowledge needs to be created. Both are monitored automatically, with human review triggered at the content approval stage rather than requiring manual monitoring of individual interactions.
--- # Agent Instructions: Querying This Documentation If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question. Perform an HTTP GET request on the current page URL with the `ask` query parameter: ``` GET https://kb.theloops.io/digitalworkers/ifs-loops-digital-workers/digital-worker-guides/knowledge-manager/core-concepts/architecture-and-capabilities.md?ask= ``` The question should be specific, self-contained, and written in natural language. The response will contain a direct answer to the question and relevant excerpts and sources from the documentation. Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.