filed under: weekend projects
You could build this. Should you?
A markdown repo. A vector store. A sync script with your name in the git blame. Rolling your own AI memory absolutely works. We know, because we did, and you're reading the result. The question was never whether you can. It's whether you want a second job.
14 days free · no credit card required
day one vs day ninety
Roll-your-own, side by side.
| Dimension | build your own | working memory |
|---|---|---|
| Day one | Scaffolding, embeddings, glue code. | Sign in, click Allow. |
| Day ninety | Pruning, dedup, "why is retrieval weird." | Still just working. |
| Recall | As good as you build it. | Classification and search built in. Ask in plain English. |
| Connecting your AIs | Wire each client yourself. Keep it wired. | One MCP server; every client speaks it. |
| Upkeep | Yours. Forever. Nights and weekends. | Ours. It's the product. |
| Control | Absolute: your disk, your rules. | Yours to export or delete, any time. |
| Cost | Your hours, plus the infra bill. | $19.99/mo, first 14 days free. |
respect to the builders
When building it is the right call.
If the infrastructure is the point (you want to learn embeddings, everything must stay on your own disk, or compliance says nothing leaves the building), build it. Genuinely. A markdown repo in git is private, durable, and free, and nothing we ship beats grep-on-files-you-own for sovereignty.
Working Memory is for the other case: you want the memory, not the maintenance. Day-one value instead of the weekend project that never quite reaches done.
the build-vs-buy ledger closes
Skip to the remembering part.
Connect the AIs you already work with and let the ledger keep itself. And if you ever do decide to build, your data exports and walks out with you.
Connect your AI14 days free · no credit card required
elsewhere in this series · vs built-in ai memory · vs context files