AI & Automation
Agent Harness
An agent harness is the execution scaffolding that connects tool calls, memory, and a repeated execution loop to an LLM, so that a single model operates as an agent that actually performs work.
Huuush · Knowledge
AI & Automation Glossary.
AI models, agents, and workflow automation.
Glossary
8 terms
AI & Automation
AGI (Artificial General Intelligence)
AGI, or artificial general intelligence, is AI that can learn and solve a wide range of problems on its own, the way people do, instead of being limited to specific tasks. There is still no agreed benchmark for reaching it or timeline for when it might happen.
AI & Automation
AI Agent
An AI agent is an autonomous system built on a large language model (LLM). You give it a goal, and it plans, calls the tools it needs, and carries out multiple steps to produce a result.
AI & Automation
Benchmarks (ARC-AGI)
A benchmark is a standard evaluation that compares AI models quantitatively by giving them the same test. ARC-AGI, proposed by François Chollet in 2019, is an abstraction and reasoning benchmark that measures the ability to generalize to new problems the model has not been trained on.
AI & Automation
LLM (Large Language Model)
An LLM (large language model) is a large neural network pretrained on vast amounts of text that generates sentences by probabilistically predicting the next token.
AI & Automation
On-Device / Local LLM
An on-device or local LLM is a language model that runs directly on the user's own device, such as a personal PC or smartphone, instead of a cloud server. Because data never leaves the device, it favors privacy and offline use.
AI & Automation
Open-Weight Model
An open-weight model is an AI model whose trained neural network weights are published, so anyone can download it, run it locally, or fine-tune it. It differs from fully open-source AI, which also releases training data and code, and redistribution or commercial use may be restricted by the license.
AI & Automation
Parameters & MoE
Parameters are the weights a model adjusts during training, and their count is a common measure of model size. MoE (Mixture of Experts) is an architecture that activates only some of the total parameters for each input, reducing compute cost.