MemoryBrain
Context, memory, instructions and projects: what is each one?
Context: what the AI can use when answering now. Memory: facts it reuses across chats. Instructions: rules you give it for answering. Projects: spaces that group a piece of work’s chats, files and instructions.
What each one does
| Feature | What it contains | When you change chats | Use it for |
|---|---|---|---|
| Context | Messages, answers and material the AI can use at that moment, up to a limit | Do not assume the whole conversation transfers | The task you are working on now |
| Memory | Facts, preferences or summaries the AI keeps about you | It can reuse them depending on the app and settings | Not repeating who you are or what you prefer |
| Instructions | Rules you write about how you want it to respond | They stay until you change them, within the scope where you set them | Setting the style and rules for answers |
| Projects | A piece of work’s chats, files and instructions | Project chats share that space. Memory depends on the app and settings | Grouping material for one piece of work and keeping it apart from others |
Where to put each thing
- A request for the current answer: write it in the chat. It becomes part of the context.
- A preference you want to reuse: it can go in memory.
- A rule you want to set: put it in instructions.
- The files and status of a long piece of work: keep them in a document within the Project. Do not rely solely on memory.
A preference can also be expressed as an instruction. The practical difference is whether you want to give an explicit rule or keep a fact the AI can reuse.
What to check before relying on memory
- A long chat is not a complete copy in memory. Context has a limit. Opening another chat does not guarantee that everything from the old one will be retrieved.
- Memory and chat history are different. Depending on the app and settings, the AI can use past chats as well as saved memories.
- Memory does not replace documents. Saved facts or summaries are not a copy of the original text you read.
- A Project does not guarantee isolated memory. Check the app’s memory settings and what your plan allows.
- Claude organises memory by topic and by project. According to its help page, sensitive topics such as health, politics and religion are left out by default.
How to keep the source of a fact
Do not assume a memory includes the original wording and its source. If you need to verify a fact, keep the message or document it came from alongside the summary. Distinguish what you said, what you read and what the AI inferred.
A memory with sources lets you keep that origin as part of the note. More: where ChatGPT gets what it remembers and how to continue a long project in a new chat.
Frequently asked questions
Does memory read my old chats?
Depending on the app and settings, the AI can use past chats as well as saved memories. Check memory settings before assuming what it can retrieve.
What goes in instructions and what in memory?
Use instructions for explicit rules about how you want it to respond. Use memory for facts and preferences you want to reuse across chats. Keep a project’s detailed status in a document.
Is a project a separate memory?
A Project groups a piece of work’s chats, files and instructions. Whether its memory is separate from other chats depends on the app and settings. Do not assume it is.
Try it
See how a memory that keeps the source of everything works, no account needed.
Sources
- OpenAI, Memory FAQ.
- Anthropic, Claude’s chat search and memory. Accessed November 8, 2026.