Zep (via Graphiti) models how facts change over time — validity windows, entity graphs, sub-200ms hybrid retrieval for production agents. ReadmeDB models what your team decided — versioned markdown files plus optional distilled memories, reachable from any MCP coding agent.
| Dimension | ReadmeDB | Zep / Graphiti |
|---|---|---|
| Primary storage | Postgres-backed markdown files + memory rows | Temporal knowledge graph + vector retrieval |
| Best job | Engineering project brain across IDE agents | Enterprise agents needing fact consistency over time |
| Temporal reasoning | File version history + timestamps | Bi-temporal edges (valid_from / valid_to) |
| Human audit trail | Git-style diffs on .md files | Graph queries + episode history |
| Coding-agent MCP | 30 tools — read/write/search/patch | REST/SDK — not IDE-native |
| Auto entity extraction | Agents write explicitly | Core ingestion pipeline |
| Graph traversal | Not built | First-class (Graphiti) |
| Self-host | Hosted SaaS | Graphiti OSS + own graph DB |
| Typical buyer | Indie hackers & small eng teams on Cursor/Claude | B2B agents, CRM, compliance-heavy workflows |
You need to answer “what did we believe about this customer on March 3?” with graph-backed provenance. Sales copilots, support agents with long histories, regulated industries — Zep’s temporal model is genuinely differentiated.
Your pain is re-explaining the repo when switching from Cursor to Claude Code. You want CLAUDE.md, decisions.md, and tasks.md as living files — not a graph DB you operate. Setup is one MCP block; humans can read the brain in the dashboard.
We added hybrid_recall (files + memories in one search) and promote_memory_to_file so scratch facts graduate into durable markdown — a lightweight answer to “facts vs documents” without running Neo4j.
{
"mcpServers": {
"readmedb": {
"url": "https://app.readmedb.com/api/mcp",
"headers": {
"Authorization": "Bearer rdb_your_key"
}
}
}
}Start free project brain →