OpenAI Custom GPTs: What They Do and What They Remember
A custom GPT is a saved ChatGPT configuration with its own instructions, knowledge files and capabilities. Here is what it can do, what it cannot hold, and where to keep the rest of your work.
Updated 4 min read
A custom GPT is a saved version of ChatGPT that you configure with your own instructions, knowledge and capabilities. Instead of typing the same rules and context into every conversation, you build a custom GPT once and open it whenever you need it. Each conversation starts with the instructions and knowledge you set, applied automatically.
Custom GPTs are useful for repeatable tasks: a writing assistant that follows your house style, a customer support helper trained on your product documentation, a coding reviewer that applies your team's standards. For work that continues across sessions and involves active projects, tasks, and decisions, the instructions field has limits — tools like fenbs keep that ongoing work outside any one AI configuration, available to Claude, ChatGPT and other connected assistants.
How a custom GPT works
You build a custom GPT in the GPT editor inside ChatGPT. The editor has four main parts: instructions (standing rules), knowledge (files you upload), capabilities (web browsing, code interpreter, image generation) and actions (connections to external APIs). When someone opens the custom GPT and starts a conversation, it reads the instructions first, then uses the knowledge files when relevant.
The custom GPT is available through ChatGPT. You can keep it private (just you), share it with a link, or publish it. For current details on sharing options and which ChatGPT plans support custom GPTs, check OpenAI's documentation (opens in a new tab) directly.
What instructions and knowledge do
Instructions are the system prompt: standing rules applied to every response. They set the tone, the scope, what to refuse, and how to format the output. They are written once and stay until you change them.
Knowledge files are documents you upload — PDFs, text files, spreadsheets — that the GPT can search when answering a question. A custom GPT for a software product might have the user manual as a knowledge file. When a user asks how to do something, the GPT searches the manual and uses what it finds.
Both instructions and knowledge are static: they stay as you set them until you update the GPT. They do not update based on what happens in conversations.
What custom GPTs cannot hold
Custom GPTs do not remember what happened in previous conversations. Each conversation starts fresh from the instructions and knowledge. If someone asked a question in a session last week, the GPT does not know that. If someone made a decision that should affect future conversations, that decision needs to go into the instructions manually.
They also do not track active work: projects in progress, tasks with a current status, decisions that were just made, or where a piece of work was left. Instructions and knowledge are good at describing a stable context. They are not designed to hold a changing, evolving work record.
This distinction matters for teams. A custom GPT can hold the reference knowledge for a product. It cannot tell you what tasks are in progress, what was decided yesterday, or what the next step in a project is.
When to use a custom GPT
Custom GPTs are a good fit for:
- Repeatable tasks that use the same rules every time: writing in a specific style, reviewing against a checklist, answering questions from a document.
- Giving a team a shared starting point: a custom GPT that everyone opens for the same type of task starts from the same instructions, reducing inconsistency.
- Tasks that need reference material: a custom GPT with product docs, a policy manual, or an FAQ library can answer from those sources without the user having to paste them in.
When you need something more
If you need your AI assistant to track a project's current state across sessions — what was done today, what was decided, what is still in progress — the instructions field is not the right place for that. Instructions are for stable rules; project state changes daily.
For ongoing work that needs to survive a session change, a tool or model change, or a team handoff, that work needs to live outside the custom GPT. A shared work record that any connected assistant can read — regardless of which model or app they are using — keeps the project history available even when you switch tools. The custom GPT and the shared work record serve different purposes and work well together.
Keep your project context outside any one AI configuration
In fenbs, the work lives in one place: projects, tasks, decisions, rules, lessons, and where things were left. ChatGPT, Claude, Cursor and other supported assistants connect over MCP and read the same record. When the instructions field of a custom GPT is not the right home for something, fenbs is. Start with a free account.
Create a free accountFree for one person · No credit card · Works with Claude, ChatGPT, Cursor and more
How to do this with fenbs
- Create a free account and add a project for the work you are doing.
- Use your custom GPT for the tasks it is good at: repeatable, rules-based tasks with stable knowledge.
- Keep active project context — what is in progress, what was decided, where you left off — in fenbs.
- Connect ChatGPT to fenbs over Integrations so it can read the project context before working.
- Save your tested custom GPT instructions as a Skill in fenbs so you have the current version outside the editor.
To compare custom GPTs with similar features in other tools, see Claude Projects vs. custom GPTs and ChatGPT Projects vs. custom GPTs. For building AI products that go further than custom GPTs, see OpenAI agents SDK.