## Learn17
- - [RAG (Retrieval-Augmented Generation)](https://parallel.ai/learn/rag-retrieval-augmented-generation)Retrieval-augmented generation (RAG) is a technique that connects a language model to an external knowledge source and grounds answers in facts retrieved at query time.
- - [Citations API](https://parallel.ai/learn/citations-api)A citations API returns citations, references to the source documents or web pages behind a result, alongside a generated answer, extracted data, or metadata, so each claim traces to its origin.
- - [Knowledge retrieval API](https://parallel.ai/learn/knowledge-retrieval-api)A knowledge retrieval API is a service that takes a query and returns the most relevant content from a knowledge source, with relevance scores and source references rather than a generated answer.
- - [Web data API](https://parallel.ai/learn/web-data-api)A web data API is an application programming interface (API) that acts as a standardized bridge between software systems across the web. You send a request, and you get structured data back without needing to know how the other system stores or builds that data.
- - [Source grounding](https://parallel.ai/learn/source-grounding)Source grounding is the practice of tying an AI model’s output to specific, verifiable external sources retrieved when a question is asked, so its claims trace back to real documents instead of training data alone.
- - [Grounding (LLM)](https://parallel.ai/learn/grounding-llm)Grounding in LLMs means the model connects each answer to specific sources instead of relying on memory from training. The academic definition of grounding states that every claim in the response can be attributed to a document in the user-specified knowledge base.
- - [AI search API](https://parallel.ai/learn/ai-search-api)An AI search API is a developer tool that returns ranked, structured web results with citations for AI models and agents to reason over.
- - [Autonomous agent](https://parallel.ai/learn/autonomous-agent)An autonomous agent is software that pursues a goal on its own, choosing its steps and taking actions to reach it. It keeps working until it meets the goal instead of waiting for a new prompt each time. It builds on what an AI agent is by adding independence and persistence.
- - [Model Context Protocol](https://parallel.ai/learn/model-context-protocol)The Model Context Protocol (MCP) is an open standard that connects AI applications to outside tools and data. An agent uses MCP to fetch current information and run actions through a single interface, so a large language model (LLM) can work beyond its training data.
- - [MCP server](https://parallel.ai/learn/mcp-server)An MCP server is a software service that gives AI models a standard way to reach tools, data, and prompts through the Model Context Protocol. It sits between the model and your APIs, databases, or files, and it translates requests into structured, model-ready responses.
- - [Company research](https://parallel.ai/learn/company-research)Company research means collecting and making sense of public and private information about a company to answer a specific question. That question might be whether to sell to it, invest in it, work for it, or compete with it. You start with the decision, then gather the facts it needs.
- - [Tool calling](https://parallel.ai/learn/tool-calling)Tool calling lets a large language model decide when to use an external tool. The model outputs the tool name and arguments as structured data, then uses the returned result to answer. Another name for this is function calling.
- - [Web search MCP](https://parallel.ai/learn/web-search-mcp)A web search MCP is a Model Context Protocol (MCP) server that gives an AI agent live web access, letting it search, fetch, and read current web pages through one standardized interface instead of a custom integration for every source.
- - [Agentic search](https://parallel.ai/learn/agentic-search)Agentic search is a pattern where an AI agent plans, runs, reads, and refines web searches in a reasoning loop, deciding what to query next based on what it finds, rather than making a single fixed retrieval call.
- - [Headless browser API](https://parallel.ai/learn/headless-browser-api)A headless browser API gives your code programmatic access to a browser that runs without a graphical user interface (GUI). It loads and renders web pages the way a real user’s browser would, then returns the finished content your code asks for.
- - [Agentic workflow](https://parallel.ai/learn/agentic-workflow)An agentic workflow is a way of working where an AI agent plans, acts, and reviews its own results without a person driving every step. The agent splits a goal into smaller tasks, calls tools to do the work, checks what comes back, and keeps going until it finishes.
- - [Browser agent](https://parallel.ai/learn/browser-agent)A browser agent pairs a large language model with a real web browser so it can act on the web the way a person would. You describe the outcome, and the agent figures out the steps. It's a type of AI agent, so it helps to understand what an AI agent is before you build one.