All terms
Agents & tools
Custom GPT
Also known as: GPT, Custom ChatGPT, GPT Store assistant
OpenAI's user-built assistants — a base GPT model plus custom instructions, knowledge files, and optional API actions, packaged as a shareable bot.
What it means
A Custom GPT is OpenAI's name for a configured ChatGPT assistant. You give it a name, a system prompt ("You are a SQL tutor that..."), upload up to 20 reference files for retrieval, and optionally define actions — REST API calls the GPT can make on the user's behalf. You publish it to the GPT Store or keep it private. Anyone with ChatGPT Plus can use it.
Under the hood, a Custom GPT is a thin wrapper: same base model (GPT-4o, GPT-5, etc. depending on era), your system prompt prepended to every conversation, your files indexed for RAG, and your actions exposed as function calls. There's no fine-tuning. Behavior comes entirely from the prompt and the retrieval over your uploaded files.
The GPT Store, launched in early 2024, was OpenAI's bid for an "App Store for AI." In practice it's mid-tier: most popular GPTs are wrappers around well-known prompts (resume writer, image generator helper, language tutor) with modest revenue sharing. The promise of a thriving developer marketplace hasn't fully materialized, but Custom GPTs remain the easiest way for non-developers to build and share an AI tool.
Compared to alternatives: Claude Projects offer similar customization with longer context and better file handling but no public marketplace. Claude Skills are markdown-file-based and developer-friendly. Custom GPTs win on distribution (every ChatGPT user can find them) and lose on portability (locked to OpenAI's ecosystem).
Example
You build a "Brand Voice Editor" Custom GPT: system prompt with your style rules, uploaded brand guidelines PDF, and an action that posts approved drafts to your CMS. Your marketing team uses it daily without ever seeing a prompt.
Why it matters
Custom GPTs democratized AI customization for non-developers — millions exist, and many companies use them as internal tools. Whether you're choosing a platform to ship an AI assistant on or just trying to understand the landscape, knowing what Custom GPTs can and can't do (no fine-tuning, no model choice, locked to OpenAI) is table stakes.