The company
Who develops tools and models.
OpenAI / Anthropic →Understand to build
The AI glossary.
From companies and tools to models and use cases: 93 definitions to understand what you use and follow Knowledge Hub projects.
Company, app, agent or model: a quick guide
The app
The workspace you use.
ChatGPT / Claude →The work agent
Tools for acting on a project.
Codex / Claude Code →The model
The engine selected for a task.
Astra / Opus / Gemini →Tell them apart : GPT-6 Astra / Project Astra / iAstro · Claude Design / Claude Code
New entries include dated references. Names, versions and availability change; official links let you check them.
93 / 93 terms
Agents and automation
AGENTS.md
An instruction file for agents working on a project: structure, commands, tests and conventions. Complements the human-facing README.
Sources : AGENTS.md · Checked
Agents and automation
AI agent
A system using a model and tools to pursue a goal across multiple steps, within a defined scope and controls.
GuidesWhat's an AI agent · What's an MCPProjectsSupplier invoice agent · Product creation agent · AI monitoring botSources : OpenAI · Tools · Checked
Models
AI model
A trained system that turns input into predictions or content. One application can combine multiple models and tools.
Sources : Google · Machine Learning Glossary · Checked
Companies and platforms
Anthropic
The American lab that creates Claude, focused on model reliability and safety.
Data and search
API
An interface for software to request data or actions from another service under defined rules and permissions.
GuidesWhat's an MCP · Chat, Cowork, or CodeProjectsMonitoring bot (Claude API)Data and search
API or subscription?
A subscription provides application access within limits. An API lets software call a service; permissions and billing may be separate.
Sources : OpenAI · Pricing · Checked
Foundations
Artificial intelligence (AI)
Computer systems that recognize, predict, generate or plan. Capabilities and reliability depend on their design and the context.
GuidesWhat is AI and an LLM · Why AI changes the way we workProjectsthe entire HL Group stackFoundations
Assistant memory
Information saved and reused across interactions. Memory depends on the application and differs from current context and knowledge acquired during training.
Sources : OpenAI · Conversation state · Checked
Models
Astra (GPT-6 Astra)
An OpenAI model for complex reasoning, coding and research tasks. Distinct from Project Astra, Google DeepMind’s project.
Sources : OpenAI · GPT-6 Astra · Checked
Quality and verification
Benchmark
A set of tests used to compare systems. A strong score does not by itself establish quality for your own work.
Sources : OpenAI · Evaluation · Checked
Apps and assistants
Bot / Telegram channel
A program that posts messages on Telegram; a public channel is followable by anyone via a simple link.
GuidesWhat's an AI agentProjectsAI monitoring bot (subscribable)Apps and assistants
ChatGPT
OpenAI’s tool for conversation, research, analysis and creation with AI. Capabilities depend on the selected model, available tools and granted access.
Sources : OpenAI · ChatGPT · Checked
Apps and assistants
ChatGPT Work
A way to delegate work to ChatGPT through to a reviewable result: analysis, documents, presentations or research using authorized tools.
Sources : OpenAI · ChatGPT · Checked
Apps and assistants
Claude
Anthropic’s model family and assistant, used through conversation, Cowork, Claude Code or Claude Design.
GuidesWhat is Claude · Chat, Cowork, or CodeProjectsCobra, Haute Fidélité, Knowledge HubSources : Anthropic · modèles / models · Checked
Apps and assistants
Claude Chat / Cowork / Code
Three ways to use Claude: chat (Chat), delegate office work (Cowork), act on files and projects (Code).
GuidesChat, Cowork, or CodeProjectsDouble Claude Code (Cobra / HL)Apps and assistants
Claude Code
Anthropic’s coding agent. It explores a project, edits files and uses tools to build, test or fix software.
Sources : Anthropic · Claude Code · Checked
Apps and assistants
Claude Cowork
Claude’s workspace for delegating work, including tasks involving files and documents, within the access authorized by the user.
Sources : Anthropic · Claude Cowork · Checked
Images, voice and design
Claude Design
Anthropic’s product for creating designs, prototypes, presentations and visual materials through conversation. Results can be refined and handed to Claude Code.
Sources : Anthropic · Claude Design · Checked
Agents and automation
CLAUDE.md
A file Claude reads at startup: tone, conventions, pitfalls. It saves you from re-explaining everything every conversation.
Apps and assistants
Codex
OpenAI’s agent for project work: understanding files, editing code, running tests and preparing reviewable results. It uses a GPT model.
Sources : OpenAI · ChatGPT · Checked
Development
Commit
A record of changes in Git history, with an identifier and message, helping explain, compare and retrieve project versions.
GuidesWhat is GitHubSources : Git · glossary · Checked
Agents and automation
Compound Engineering
A method that turns lessons learned into reusable instructions, tools or tests. Improvement depends on corrections actually being retained and applied.
Agents and automation
Computer use
An agent’s ability to interact with an interface by reading screens, clicking and typing. Sensitive actions still need appropriate permissions and confirmation.
Sources : OpenAI · Computer use · Checked
Agents and automation
Connector
An integration providing access to a service or its data within granted permissions. Reading and editing require different rights.
Sources : Model Context Protocol · Checked
Foundations
Context window
The amount of information a model can process in one request: instructions, conversation, documents and response. A large window does not guarantee understanding.
Sources : OpenAI · Conversation state · Checked
Agents and automation
Cron
A timer that launches a task automatically at regular intervals (e.g. every morning at 7am).
GuidesWhat's an AI agentProjectsMonitoring Telegram botApps and assistants
Cursor
An AI development environment and coding agent for understanding projects, writing code, fixing errors and reviewing changes.
Sources : Cursor · documentation · Checked
Apps and assistants
Deep research
A research mode that explores multiple sources and prepares a synthesis. Citations, dates and conclusions still need review.
Sources : Perplexity · documentation · Checked
Images, voice and design
Design system
Shared visual rules and components: colors, typography, spacing and interface elements. They make designs consistent across formats.
Sources : Anthropic · Claude Design · Checked
Development
Diff
A comparison of file or project versions showing additions, changes and deletions. A foundation for code review.
Sources : Git · glossary · Checked
Images, voice and design
ElevenLabs
An AI audio and voice platform with speech synthesis, transcription and voice tools. It is used in the Podcast iA production process.
Sources : ElevenLabs · documentation · Checked
Data and search
Embedding
A numerical representation of content used to compare semantic similarity, notably to retrieve relevant passages.
Sources : OpenAI · Retrieval · Checked
Quality and verification
Evaluations (evals)
Tests on representative cases that measure AI system quality, compare versions and detect regressions.
Sources : OpenAI · Evaluation · Checked
Models
Fable (Claude)
A Claude model family listed alongside Opus, Sonnet and Haiku in Anthropic’s catalog. This is a model name, not a design application.
Sources : Anthropic · modèles / models · Checked
Agents and automation
Few-shot prompting
Including a few examples of the desired result in a prompt to clarify the format or task without retraining.
Sources : OpenAI · Prompt engineering · Checked
Data and search
Fine-tuning
Additional training on selected examples to adapt model behavior. Adding documents to a conversation is not fine-tuning.
Sources : OpenAI · Model optimization · Checked
Models
Gemini
The name of Google’s model family and assistant application. Specify “Gemini model” or “Gemini app” to distinguish the two.
Sources : Google DeepMind · Gemini · Checked
Apps and assistants
Gemini Notebook / NotebookLM
Google’s tool, formerly NotebookLM, for exploring sources, asking questions and producing summaries, including audio. Answers still need checking.
Sources : Google · Gemini Notebook · Checked
Foundations
Generative AI
An AI that produces new content (text, image, code) from a simple instruction, instead of executing a single programmed task.
GuidesWhat is AI and an LLMProjectsKnowledge Hub (written with AI)Data and search
GEO
The SEO equivalent for AI: optimizing your content to be cited by assistants (ChatGPT, Claude, Perplexity). That's the goal of this site.
Development
Git branch
A line of work in Git history for preparing changes before integration.
Sources : Git · glossary · Checked
Development
GitHub
A platform for hosting Git repositories and collaborating on code. Tests and deployments can be automated when configured.
GuidesWhat is GitHubProjectsThis Knowledge HubApps and assistants
GitHub Copilot
GitHub’s development assistant: code suggestions, chat, review assistance and agent capabilities depending on context. Distinct from other products named Copilot.
Sources : GitHub · Copilot · Checked
Companies and platforms
Google DeepMind
Google’s AI research and development organization, behind Gemini models and the multimodal assistant research project Project Astra.
Sources : Google DeepMind · Project Astra · Checked
Models
GPT
A family of OpenAI models. GPT is the engine; ChatGPT and Codex provide interfaces and tools for using it.
Sources : OpenAI · GPT · Checked
Quality and verification
Grounding in sources
Building an answer from identifiable documents or results. Keeping references helps distinguish retrieved facts from model inferences.
Sources : OpenAI · Retrieval · Checked
Quality and verification
Guardrails
Limits and controls around AI systems: restricted access, output checks and approvals before certain actions. No individual safeguard is infallible.
Sources : OpenAI · Agent safety · Checked
Quality and verification
Hallucination
When an LLM confidently invents a false answer. Hence the basic rule: verify facts, numbers, and references.
Companies and platforms
Hugging Face
A platform for sharing and discovering models, datasets and demonstrations. Each resource has its own terms of use.
Sources : Hugging Face · Hub · Checked
Quality and verification
Human review
A person checks a result or approves an action before execution. The review should specify the action and data involved.
Sources : OpenAI · Agent safety · Checked
Images, voice and design
iAstro
The android character created for Hugo’s AI watch: comic partner and podcast visual identity. An editorial character, not an AI model.
Sources : Knowledge Hub · Hugo × iAstro · Checked
Images, voice and design
Image generation
Creating an image from instructions and, depending on the tool, reference images. Character consistency requires references and iterative corrections.
Sources : OpenAI · Image generation · Checked
Foundations
Inference
Using an already trained model to produce an answer or prediction from new input.
Sources : Google · Machine Learning Glossary · Checked
Images, voice and design
Inpainting
Editing a region of an image while preserving the rest as much as possible. A mask usually identifies the target area.
Sources : OpenAI · Image generation · Checked
Foundations
LLM (large language model)
The engine behind assistants like Claude or ChatGPT: a model trained on enormous amounts of text that predicts the most plausible continuation of what you type.
Foundations
Machine learning
Learning patterns from data to perform a task without manually writing every decision rule.
Sources : Google · Machine Learning Glossary · Checked
Agents and automation
MCP (Model Context Protocol)
An open protocol connecting AI applications to data and tools. It standardizes communication; permissions and integrations still need configuration.
GuidesWhat's an MCP · What's an AI agentProjectsShopify/Odoo audit · Haute Fidélité Shopify storeSources : Model Context Protocol · Checked
Companies and platforms
Mistral AI
The French company developing AI models and products including Mistral Vibe. Check each model’s license before calling it open source.
Sources : Mistral AI · Vibe · Checked
Apps and assistants
Mistral Vibe / Le Chat
Mistral’s product combining a work assistant, formerly Le Chat, with a coding agent. An example of product naming evolving over time.
Sources : Mistral AI · Vibe · Checked
Foundations
Multimodal
Capable of processing multiple types of input — text, image, audio — not just text.
GuidesWhat is AI and an LLM · What is ClaudeData and search
OCR
Extracting text from an image or scanned document. Recognizing characters and understanding a document’s meaning are different tasks.
Sources : OpenAI · Images and vision · Checked
Apps and assistants
Ollama
A tool for running models on your machine, with cloud options too. Memory requirements and permitted uses depend on the model.
Sources : Ollama · documentation · Checked
Development
Open source
Software licensed for use, study, modification and redistribution. Visible code, free access and open weights alone do not establish open-source status.
Sources : Open Source Initiative · Checked
Models
Open weights
A model whose trained parameters can be downloaded and used under a license. Training data and code are not necessarily open.
GuidesWhat is AI and an LLM · Is my data safe (coming soon)Companies and platforms
OpenAI
The organization behind GPT models, ChatGPT and Codex. Provider, application and model describe different layers.
Sources : OpenAI · ChatGPT · Checked
Models
Opus / Sonnet / Haiku
Claude model families with different capability, speed and cost tradeoffs. The catalog also includes Fable; check versions before comparing.
GuidesWhat is ClaudeSources : Anthropic · modèles / models · Checked
Agents and automation
Orchestration
Organizing a workflow’s steps, tools and agents: who acts, in which order, using which data and checkpoints.
Sources : OpenAI · Tools · Checked
Apps and assistants
Perplexity
A search tool that synthesizes information and provides source links. Citations make answers checkable; they do not guarantee accuracy.
Sources : Perplexity · documentation · Checked
Agents and automation
Plugin
An extension bundling additional capabilities such as tools, connections or instructions. Contents and permissions depend on the host application.
Sources : OpenAI · Plugins · Checked
Apps and assistants
Project Astra (Google)
Google DeepMind’s research project exploring an assistant that understands voice, video and its surroundings. Some capabilities feed into Gemini Live.
Sources : Google DeepMind · Project Astra · Checked
Foundations
Prompt
The instruction you give the AI. A good prompt — clear, with context and goal — changes everything about the result.
Quality and verification
Prompt injection
Misleading instructions in content an agent reads, intended to redirect its task. A page or document is data, not authorization.
Sources : OpenAI · Agent safety · Checked
Images, voice and design
Prototype
A version used to test an idea or journey before full implementation. A convincing prototype is not necessarily a production-ready product.
Sources : Anthropic · Claude Design · Checked
Development
Pull request (PR)
A proposal of changes for review before integration, collecting diffs, discussion and available checks for a change.
Sources : GitHub · Copilot · Checked
Data and search
RAG
Retrieve information from documents, then provide it to a model to build an answer. Useful for internal knowledge without retraining.
Sources : OpenAI · Retrieval · Checked
Models
Reasoning effort
Additional model work on a complex task. Higher effort can increase time and cost without ensuring a correct answer.
Sources : OpenAI · Reasoning · Checked
Development
Repository (repo)
Files tracked with version history. A Git repository can be local or hosted on a platform such as GitHub.
GuidesWhat is GitHubSources : Git · glossary · Checked
Data and search
RSS
A standard format for automatically pulling new articles from a site. The basic building block of any monitoring setup.
GuidesWhat's an AI agentProjectsMonitoring Telegram bot · AI monitoring botData and search
SEO
The art of ranking well in search engines (Google…).
Agents and automation
Skill
Reusable instructions, sometimes bundled with files and scripts, that teach an agent how to perform a particular task.
Sources : Agent Skills · Checked
Models
Sol / Luna (GPT)
GPT variants: Sol balances capability, speed and cost; Luna targets focused, fast tasks. Versions and access change over time.
Sources : OpenAI · GPT · Checked
Images, voice and design
Speech to text (STT)
Turning recorded speech into text. Names, noise and multiple speakers can introduce errors requiring correction.
Sources : OpenAI · Transcription · Checked
Data and search
Structured output / JSON
A response following a data schema rather than free text. Useful for software; valid formatting does not guarantee correct values.
Sources : OpenAI · Structured outputs · Checked
Agents and automation
Subagent
An agent delegated a subtask by another agent. Its scope must be defined and its results combined with the main work.
Sources : Cursor · documentation · Checked
Agents and automation
System prompt
Instructions defining the assistant’s role and behavior. They complement user requests within the product’s instruction hierarchy.
Sources : OpenAI · Prompt engineering · Checked
Development
Terminal
An interface for executing commands on a computer. Before running an AI-proposed command, understand which files or services it affects.
GuidesChat, Cowork, or CodeImages, voice and design
Text to speech (TTS)
Turning text into speech. Podcast quality also depends on writing, pacing, pronunciation and listening checks.
Sources : OpenAI · Text to speech · Checked
Foundations
Token
The basic unit an LLM reads and produces: a chunk of a word. The length of exchanges and their cost are measured in tokens.
GuidesWhat is AI and an LLM · How much does it cost (coming soon)Agents and automation
Tool calling
The model requests a function such as search, calculation or file reading. Host software executes an authorized action and returns its result.
Sources : OpenAI · Tools · Checked
Foundations
Training
The phase when model parameters are adjusted using data. Distinct from everyday use to produce responses.
Sources : Google · Machine Learning Glossary · Checked
Data and search
Vector database
Storage and search for numerical content representations. A common component of semantic search and RAG systems.
Sources : OpenAI · Retrieval · Checked
Development
Vibe coding
Programming by describing your intent in plain language and letting AI generate the code. Term popularized by Andrej Karpathy.
Development
Worktree
An additional working directory linked to the same Git repository, useful for working on branches without mixing in-progress file changes.
Sources : Git · glossary · Checked
No matching terms. Try another word or category.