1. Using the “Knowledge Retrieval Node” to Access the Knowledge Base
The Knowledge Retrieval Node is one of the core capabilities in XpertAI agent workflows. It can automatically search the constructed knowledge base for content fragments most semantically relevant to the user’s input question and pass these fragments as contextual information to downstream agent nodes. This process ensures that the agent’s responses are always based on real knowledge sources, rather than solely generated by language models, significantly improving the accuracy and consistency of answers. Typical use cases:- Providing factual support for LLMs in Q&A workflows;
- Supplying industry or internal company background knowledge for agents in business analysis or report interpretation;
- Serving as a knowledge pre-processing module in complex tasks such as document summarization and decision support.
2. Empowering Agents to Autonomously Access the Knowledge Base via “Knowledge Base Tool Node”
Beyond fixed retrieval nodes, XpertAI also supports registering the knowledge base as a Tool for agents. In this mode, agents can autonomously decide whether to access the knowledge base during conversations or task execution, dynamically integrating retrieval results into their responses. You can add a “Knowledgebase Node” to the agent— this node allows the agent to proactively access the knowledge base to supplement knowledge or verify information as needed. For each agent-to-knowledge-base binding, you can configure both:- a fixed filter that limits the domain, project, folder, or document status and cannot be changed by the agent;
- Allow agent automatic filtering, which lets the agent use file names, file types, folders, and metadata to further narrow the current request;
- a live knowledge filter options tool, which lets the agent inspect valid folders and metadata values instead of guessing them;
- a knowledge graph explorer when GraphRAG is enabled, allowing the agent to find entities, follow relationships, inspect bounded evidence, and then turn the discovered terms into a source-chunk retrieval;
- Vector retrieval, with intelligent filtering fully supported on PGVector and Milvus.
- Enhances agent autonomy and contextual flexibility;
- Suitable for long conversations and multi-turn reasoning in complex agents;
3. Enabling a General Knowledge Base in the Project Workspace
For collaborative scenarios across workflows or multiple tasks, XpertAI supports adding a general knowledge base in the Project Workspace. This knowledge base is available to all general agents within the project, allowing them to automatically retrieve required knowledge when performing any task, without additional node configuration. This enables your project team to:- Consolidate internal documents, operation manuals, industry reports, etc., into a unified knowledge source;
- Allow agents to directly reference shared knowledge when performing report analysis, customer Q&A, document writing, and other tasks;
- Ensure all team agents work based on the same information system.
4. Sharing one knowledge base with fixed and agent-generated filters
When several teams share a large knowledge base, you do not need to duplicate the documents for every domain. Configure a different fixed filter on each agent or workflow binding, then enable agent filtering where it adds value. For example:- a water-resources quoting agent is fixed to
metadata.domain = water-resources; - a logistics quoting agent is fixed to
metadata.domain = logistics; - each agent can add a year, region, file type, or file-name keyword found in the user’s question;
- all results remain subject to organization permissions, disabled-content checks, and the fixed boundary.