๐๐น๐ฎ๐๐ฑ๐ฒ ๐๐๐ถ๐น๐ฑ๐ ๐ ๐ ๐ฒ๐บ๐ผ๐ฟ๐ ๐ช๐ถ๐ธ๐ถ ๐๐ผ๐ฟ ๐ฌ๐ผ๐
Work with Claude on one project for weeks. It remembers your decisions. It knows your style. You do not need to explain things twice.
This memory is not a complex database. It is a folder of plain text files. Andrej Karpathy calls this an LLM Wiki.
The system is simple:
- One index file lists every note.
- Each note holds one specific fact.
- Notes link to other notes.
Two habits make this work:
- Keep it lean. Small files save space.
- Use rules. Claude updates old notes instead of making new ones.
This keeps the memory clean. Knowledge builds up. The AI edits facts instead of starting over.
This works for one person. It works for a few hundred notes. Large teams need bigger tools like vector databases. For personal work, plain files are enough.
You do not need to set this up. Claude does it for you.
Try this: Work on a real project for a few weeks. Open the memory folder. The structure is there.
Source: https://dev.to/ankushchadha/use-claude-long-enough-and-youll-end-up-with-karpathys-llm-wiki-without-doing-much-3cmo Optional learning community: https://t.me/GyaanSetuAi