Why AI memory belongs in Markdown you own.
Most memory tools make you pick: a database your AI can query but you can't read, or a folder you can read but your AI can't really use. We built both. Plain files, fully indexed.
## Why plain text matters
AI agents need context, and context has to live somewhere. If it only lives inside one product's database, it's fragile. If it lives in ordinary files, you can read it, edit it, version it, move it, and keep it.
Your memory is plain Markdown, not embeddings in a store you can't open. Open any file and read exactly what your AI knows. In English, not float vectors.
Files on your disk or in your private cloud. Version them in Git, back them up, export them, or self-host the whole thing. No vendor can take them away.
Formats outlast products. A folder of Markdown will still open in ten years, long after this season's AI memory startups have pivoted or shut down.
## The catch
No. And this is where most plain-text setups stop. A folder of files is readable and yours, but an AI can only grep it or reload whole files. There's no semantic search, no graph, no way to filter by field, and it falls apart at scale. The files are the easy half. Making them queryable without giving up the plain files underneath is the hard half.
## The synthesis
Basic Memory keeps the plain Markdown as the source of truth and builds a real index on top of it: full-text search, semantic (vector) search, a knowledge graph of typed relations, and schemas that validate note types. Read it like a doc; query it like a database.
| Capability | Basic Memory | Vector store | Plain files |
|---|---|---|---|
| You can read and edit it | |||
| Full-text + semantic search | |||
| Knowledge graph + relations | |||
| Structured schemas | |||
| Survives if the vendor disappears | |||
| Works across every AI tool (MCP) |
## Together with built-in memory
Built-in memory in Claude or ChatGPT is good at short-term. Your preferences, what you talked about yesterday, the working context of a thread. Keep it on. We're the long-term layer underneath: owned, structured, and large enough to hold a real second brain. They're built to work together.
## Validation
Kind of. In June 2026, Google published the Open Knowledge Format. organizational knowledge as a directory of Markdown files with frontmatter and links. It's a tiny industry sigh of relief: yes, this is the shape that lasts.
“The answer to this problem isn't another knowledge service. You need a format.”— Google Cloud, on the Open Knowledge Format
Exactly. The format is the easy part. And now everyone agrees on it. The hard part is making those files actually usable. That's what Basic Memory has done since 2024. Read the full take →
Plain Markdown you own, indexed so your AI can actually use it. Free and open source to run locally.
Knowledge tools built to last
© 2026 Basic Memory - Basic Machines