This open source package lets you store your memories on @github
It's super customizable:
🤖 Instruct the agent when to save memories
✍ Use your own memory writing prompt
🏷️ Add custom frontmatter
✅ Organize memories however you want
🎯 Scope memories to specific directories to avoid context pollution
you could also give memory.md files a custom scope too (it normally defaults to the parent package)
matching goes up and down the path. Search looks through parents that include your folder, plus children under it
https://t.co/WmfFFAk3WV
why don't you store your agent's memories in git?
this is how tiramisu picks what the agent searches. you’re in apps/api, so it gets root + apps/api and ignores apps/web + apps/mobile
personal/team/org repos can opt in with availableToWorkspace 🧵
@simplifyinAI I do the same thing with curated markdown in the repo, and the agent uses it when the task needs it. Open sourced as tiramisu: https://t.co/xmu94Vg3TD
@its_LtDan Yeah CLAUDE.md for rules, but decisions / common errors shouldn’t always be in context. Better as memory you load when needed. You might want to try tiramisu on npm
@grunersoftware Hey Martin. Happy to connect. It doesn’t automatically commit the files. I think the nicer flow is to allow you or your agent to commit them alongside the code.
@rexan_wong Half agree. One place agents can read and update, yes. If that place is Notion, you’re still outside the code. I keep decisions and gotchas as markdown in the repo so the coding agent already has them. tiramisu if useful: https://t.co/xmu94Vg3TD
@TipsCsharp Agree on new chat. Disagree that pasting three lines is enough once the project has decisions and gotchas. That’s when you need memories in the repo you can search, not a summary you rewrite by hand. That’s what tiramisu is: https://t.co/xmu94Vg3TD