The Markdown File That Beat a $50M Vector Database | by Micheal Lanham
Short summary
Three major AI agent platforms—Manus, OpenClaw, and Claude Code—store their memory in plain Markdown files instead of vector databases. The article breaks down how file-based context boosts token-cache economics, enables attention control, and layers optional semantic retrieval, plus when this approach starts to break down.
- Manus hit $100M ARR in eight months using append-only Markdown for memory instead of a vector database, since its 100:1 input-to-output token ratio makes cache-friendly file appends roughly 10x cheaper than invalidating a cache with database writes.
- OpenClaw layers SQLite-based vector search directly on top of its Markdown notes (vectorWeight = 0.7 plus temporal decay) rather than using an external vector database, blending keyword and embedding scores.
- Claude Code's CLAUDE.md files use directory/scope-based progressive disclosure to load only relevant context, replacing the need for a retrieval database entirely.
- All three systems treat the filesystem hierarchy itself as the retrieval mechanism, using plain text files for both long-term memory and active attention control (e.g. Manus's constantly rewritten todo.md).
Full summary
Three leading AI agents—Manus, OpenClaw, and Claude Code—all use plain Markdown files as their main memory layer instead of expensive vector databases. Manus hit $100 million ARR in eight months and attributes its design to cost savings: it processes 100 input tokens for every output token, so caching input via append-only Markdown slices input costs by tenfold. Their todo.md file does double duty—storing the plan and constantly rewriting the most recent steps so the model’s attention stays on track during long, multi-step tasks.
OpenClaw starts with a simple file layout—one MEMORY.md for core facts and daily notes in dated files—and layers on lightweight semantic search. It compacts old chat history by having the agent write out key memories before truncation, then runs SQLite-based vector queries over those same Markdown files. There’s no external service. The vector index lives alongside the source files and can blend keyword hits with embedding scores, using parameters like vectorWeight = 0.7 and temporal decay so stale notes don’t crowd out fresh ones.
Anthropic’s Claude Code embeds the Markdown approach into its official product. CLAUDE.md files hold everything from company policies to project-specific workflows. The system only loads files relevant to the current directory or user scope, cutting down on context bloat. That progressive disclosure replaces the need for a database-driven retriever: the filesystem’s hierarchy drives what the model reads when a session starts.