Reference
AI Glossary
A concise reference for the terms used across AI systems.
Agentic AIAI systems designed to plan or act through tools toward a defined goal, with controls and review appropriate to the task.AgnoAn open-source framework used to build AI agents and workflows around language models.AI AgentSoftware that uses an AI model to select or perform steps toward a goal, often with tools, data, or human approval.Bielik AIA family of Polish-language language models developed by the SpeakLeash community.ChatGPTA conversational AI product from OpenAI that uses generative language models.ChromaDBAn open-source database for storing and retrieving embeddings and associated metadata.ClaudeA family of AI models and products developed by Anthropic.Claude CodeAn Anthropic coding tool that can work with a codebase through a command-line interface.CrewAIA framework for coordinating multiple AI agents around shared tasks and workflows.DeepSeekAn AI company that develops language models and related tools.ElevenLabsA company that provides speech synthesis, voice, and audio-generation software.EmbeddingA numeric representation of data, such as text or images, used to compare meaning or similarity.Fine-tuningFurther training of a pre-trained model on selected examples to adapt its behavior to a task or domain.GeminiA family of AI models and products developed by Google.Gemini 3 FlashA Gemini model name used in this glossary; availability and technical specifications should be checked in current Google documentation.Gemini 3 ProA Gemini model name used in this glossary; availability and technical specifications should be checked in current Google documentation.GPT-5A GPT model name used in this glossary; availability and technical specifications should be checked in current OpenAI documentation.GPT-5.2A GPT model name used in this glossary; availability and technical specifications should be checked in current OpenAI documentation.GPT CodexA coding-oriented GPT model or product name; its capabilities depend on the current OpenAI release and configuration.GrokA family of AI models and products developed by xAI.Haiku 4.5A Claude model name used in this glossary; availability and technical specifications should be checked in current Anthropic documentation.HallucinationAn AI output that is inaccurate, unsupported, or invented while appearing plausible.Input TokenA unit of model input used when text, code, or other data is processed by a language model.LangChainAn open-source framework for building applications that use language models, tools, and external data.LangGraphA LangChain ecosystem library for representing agent workflows as graphs with states and transitions.LangSmithA platform for tracing, evaluating, and monitoring language-model applications.LatencyThe time between a request and the response or result it produces.LlamaIndexA framework for connecting language models to external data sources and retrieval workflows.LLMLarge language model: a machine-learning model trained to process and generate language.ManusAn AI agent product name; its features and availability should be checked with the provider.Nano-BananaAn AI model or product name used in this glossary; its provider and specifications should be confirmed in current documentation.OllamaSoftware for running and managing selected language models locally or in self-hosted environments.OpenAIAn AI research and product company that develops models, tools, and services.Output TokenA unit of model-generated output used when a language model returns text or code.PineconeA managed database service designed for vector search and retrieval.PydanticAIA Python framework for building typed applications that use language models.QdrantAn open-source vector database for similarity search and retrieval.RAGRetrieval-augmented generation: a pattern that retrieves relevant data before a model produces an answer.Sonnet 4.5A Claude model name used in this glossary; availability and technical specifications should be checked in current Anthropic documentation.SoraAn OpenAI product name associated with generative video; features may vary by release and region.Speech-to-TextTechnology that converts spoken audio into written text.Structured OutputA model response constrained to a defined schema or data format, such as JSON.text-embedding-3An OpenAI embedding model family name; available variants and specifications should be checked in current documentation.Text-to-SpeechTechnology that converts written text into synthetic speech.Vector DatabaseA database designed to store vectors and find records that are mathematically similar to a query vector.Vercel AI SDKA software development kit from Vercel for adding AI-model interactions to web applications.Voice AIAI systems that analyze, generate, or interact through spoken audio.xAIAn AI company that develops models and products including the Grok family.Transformer ArchitectureA neural-network architecture that uses attention to process relationships within sequences of data.Attention MechanismA model component that weighs the relevance of different parts of an input while producing an output.TokenA unit into which model input and output are divided for processing and accounting.Context WindowThe maximum amount of input and prior conversation a model can consider in one request.TemperatureA generation setting that influences how predictable or varied a model response is.Prompt EngineeringThe practice of designing instructions, context, and examples to guide a model toward a useful response.Function CallingA mechanism through which a model requests a defined software action with structured arguments.Zero-Shot LearningPerforming a task using an instruction without providing task-specific examples in the prompt.Few-Shot LearningGuiding a model with a small number of examples before asking it to perform a similar task.Model ParametersValues learned during model training that influence how the model transforms inputs into outputs.Hugging FaceA company and software platform that hosts machine-learning models, datasets, and related tools.AutoGenA framework for building multi-agent applications and automated workflows around language models.HaystackAn open-source framework for retrieval, question answering, and language-model pipelines.LiteLLMA software library that provides a common interface for calling multiple language-model providers.Mistral AIAn AI company that develops language models and related products.CohereAn AI company that develops language models and enterprise-focused AI products.ReplicateA platform for running and deploying machine-learning models through hosted infrastructure and APIs.RLHFReinforcement learning from human feedback: a training approach that uses human preference data to shape model behavior.LoRALow-rank adaptation: a method for adapting a model by training a small set of additional parameters.QuantizationA technique that represents model values with lower numerical precision to reduce memory use or speed inference.