The Future Tech Blog focuses on tools that provide clear utility. Today, we're introducing query, an AI agent skill designed to help developers and teams extract grounded information from their knowledge bases. It’s built to answer questions by thoroughly searching your brain knowledge base, returning answers complete with citations. This tool efficiently handles tasks like complex lookups, background research, and mapping relationships within your data. The goal of query is to provide factual, traceable answers based solely on your internal information, avoiding unsupported claims.
How Query Works
When you ask query a question, it follows a structured, five-phase process to deliver an answer. First, it takes your initial question and decomposes it into several distinct search strategies. This breakdown ensures that different aspects of your query are addressed systematically.
Next, query runs hybrid searches. This means it combines both keyword-based searches and semantic searches to find relevant information within your knowledge base. This dual approach helps uncover both direct matches and conceptually related content.
In the third phase, after identifying initial results, the tool reads the top results in full when needed. This deep reading ensures that the context and details of the information are correctly understood, moving beyond just snippets.
Following the reading phase, query synthesizes its findings. This involves compiling the gathered information into a coherent answer. Critically, every claim made in the answer is accompanied by specific citations, tracing back to its source within your knowledge base.
Finally, the process concludes by flagging any information gaps. If query identifies that a part of your question cannot be fully answered with the available data, it will indicate where the knowledge is insufficient.
Practical Applications
query excels at providing answers by drawing directly from your organization's brain knowledge base. Its primary function is to return grounded answers with citations. This means it’s not for answering from general knowledge when the brain has relevant content. The tool strictly prohibits hallucination, ensuring that all information presented is verifiable within your supplied data.
Common triggers that activate the query skill include phrases such as 'what do we know about', 'tell me about', 'who is', 'search for', and 'connections'. For example, if you ask, 'what do we know about the Q3 project budget allocations in Europe?', query will analyze the question, search your knowledge base for relevant documents, and present a summarized answer with references to the specific pages or documents where the budget data is stored.
The tool is specifically designed for scenarios requiring deep dives into proprietary information, whether it's understanding historical project decisions, researching internal company policies, or mapping relationships between different departments or initiatives. It acts as a dedicated research assistant for your internal data.
Query Modes and Source Hierarchy
query operates through several distinct modes to perform its searches and analyses. These include keyword search, which identifies exact or near-exact term matches; hybrid semantic search, which understands the meaning and context of your query; full-page retrieval, for when an entire document needs to be read to extract details; backlink lookup, to find all references to a specific piece of information; timeline queries, to retrieve chronological data; and graph traversal, specifically for answering relationship questions by navigating connections within your knowledge graph.
When synthesizing answers, query respects a clear source hierarchy to prioritize information accuracy and relevance. User statements rank highest, meaning direct input or clarifications from you take precedence. Next in line is compiled truth, representing verified or aggregated data within the knowledge base. Timelines follow, providing chronological context. External sources, while valuable, rank lowest in this hierarchy. This structured approach ensures that the most authoritative and relevant information is always surfaced first. Every answer traces claims to specific page slugs, reinforcing its grounded nature.
FAQ
Q: Can query answer questions about general topics like history or science? A: No, query is designed to answer questions by searching your specific brain knowledge base. It avoids using general knowledge when your brain has relevant content.
Q: What happens if there isn't enough information in the knowledge base to answer my question? A: query will flag information gaps if it cannot fully answer your question with the available data, indicating where the knowledge is insufficient.
Q: How does query prevent making up answers? A: The tool strictly prohibits hallucination and is engineered to return grounded answers with citations, tracing every claim to specific sources within your knowledge base.
This agent skill offers a way to get directly cited answers from your internal information. Its structured approach helps teams quickly find specific data points and understand relationships within their knowledge base.
