Concept L2 · Context engineering informational

Memory Isn't a Knowledge Base — Use Projects for That

What it is

A boundary rule: Memory is for a small set of durable facts about you, not a store for documents or structured reference material — that belongs in a Project's files.

structured reference →large volume ↑Project filesDocuments, specs, datasets youreferenceProject filesGrowing reference materialMemoryA handful of durable personalfactsMemorySingle preferences andconstraints
Memory holds a few facts; documents belong in a Project.

Why it works

People try to make Memory hold everything, but it's designed for lightweight, persistent facts, not a searchable document store. Cram reference material in and it becomes bloated and unreliable. Projects exist for exactly that: files and instructions scoped to a body of work, retrieved when relevant. Using each for its purpose keeps both effective.

When to use it

When you're deciding where to put reference material you want ChatGPT to use repeatedly — notice whether it's a fact or a document.

When not to use it

For genuine single facts ('I prefer concise answers'), Memory is right — don't over-engineer a Project for one preference.

Prompt

I want ChatGPT to reliably use <this information> in future work. Is this a Memory fact or Project reference material? Assume Memory is for a few durable personal facts and Projects hold documents and specs.

Example

Your writing style preference → Memory; your 40-page brand guidelines PDF → a Project's files, where ChatGPT can actually reference the detail.

Common mistakes

  • Pasting long documents into Memory and expecting reliable recall.
  • Bloating Memory until the genuinely useful facts get crowded out.
  • Not using Projects for reference material that a Project handles far better.

Related