Datailor is a solution to a personal frustration. Session memory is a dead end. Teach Claude your coding style today, start a new thread tomorrow, and you're explaining indentation preferences all over again. AGENTS.md promises permanence, but it's a static wall of text that never learns, never adapts, and quietly becomes outdated the moment your workflow shifts. LLM-native memory is opaque, unreliable, and hopelessly vendor-locked. Cross-device portability? Try migrating your "preferences" from Cursor to Claude to Codex and watch them evaporate. I wanted something that actually understands the problem: ◼ Dynamic matching — knows when to surface a preference and when to stay out of the way ◼ Automatic capture — learns from what you do, not from what you claim you do ◼ Intentional expiration — lets stale preferences fade instead of accumulating digital hoard ◼ Conflict detection — surfaces contradictions instead of silently guessing wrong So I built Datailor. One Markdown file. Any agent can read it. Datailor is deliberately minimal, human-readable, and open source. I beleive AI tooling should be like good conversations: respectful, interactive, and actually listening.

Datailor Preference MCP

Give every AI agent your defaults

Give every AI agent your defaults