Glossary
Definitions for the next generation of AI-native documentation.
MCP (Model Context Protocol)
An open standard that enables AI models to interact with external tools and data sources. SotaDocs provides an MCP server to deliver documentation context directly to agents like Claude and Cursor.
Context Vault
A secure, centralized repository in SotaDocs that stores semantic indexes and documentation snapshots, allowing AI agents to maintain consistent knowledge across different sessions.
Skills.md
A project-specific instruction file (e.g., GEMINI.md, CLAUDE.md) generated by SotaDocs. It provides AI agents with instant context about a technology stack, reducing the need for repetitive per-request documentation retrieval.
Blindspot Detection
The automated process of identifying gaps between an AI model's training data cutoff and the current state of documentation. SotaDocs flags these blindspots to prevent model hallucinations.
Summary Mode
A retrieval mode that provides a high-level, token-efficient overview of documentation. Ideal for giving agents a broad understanding of a library or API without consuming excessive context window space.
Vector Search
A retrieval technique that uses semantic embeddings to find the most relevant pieces of documentation for a specific natural language query. It enables precise, cited answers from AI agents.
Hallucination (AI)
A phenomenon where an AI model generates factually incorrect or nonsensical information. In coding, this often happens when an agent guesses API signatures because it lacks up-to-date documentation.
RAG (Retrieval-Augmented Generation)
A technique that enhances LLM responses by retrieving relevant information from an external knowledge base (like SotaDocs) before generating a final answer.
Context Window
The maximum amount of text (measured in tokens) an AI model can process in a single request. SotaDocs optimizes window usage by filtering out irrelevant documentation.
Token Efficiency
The practice of reducing the number of tokens sent to an LLM to lower costs and latency while maintaining or improving response quality.
Agentic IDE
An Integrated Development Environment (like Cursor, Windsurf, or PearAI) built to support AI agents that can autonomously read, write, and debug code.
Version-Awareness
The capability of a documentation engine to retrieve context that matches the exact version of a library or API used in a project's local environment.
Lockfile Detection
The automated process of scanning dependency manifests (e.g., package-lock.json, poetry.lock) to identify precise library versions for context retrieval.
Prompt Patching
A technique where targeted documentation updates are injected into an agent's prompt to correct specific knowledge gaps or 'patch' detected blindspots.
Citations
Verifiable source links, line numbers, and timestamps included with retrieved context to help agents attribute information and developers verify accuracy.
MCP Tool
An executable function exposed via the Model Context Protocol (like a documentation search) that an AI agent can call to perform actions or fetch data.
MCP Resource
A static data object (like a Markdown file or schema) that an AI agent can read via the Model Context Protocol to gain specific background information.
Semantic Indexing
The process of converting raw documentation into mathematical vectors (embeddings), enabling agents to search for context by meaning rather than simple keywords.