remindy docs

Linters catch syntax. remindy catches taste: AI-slop copy, bespoke UI, vague commits, security foot-guns, features nobody asked for.

Better models won’t fix it: they still don’t know how you work, and forget it when you switch tools. Teach remindy once; it holds every agent to your standards before it writes.

Portable via MCP

A correction made in one tool (say Cursor) is respected by another (Kiro, Claude, Windsurf). One shared memory, every agent.

About 15 tokens a turn

Each rule compresses to one caveman line, about 15 tokens. Recall ranks locally with no model call, and injects only the few rules that match.

Runs on your machine

Storage and ranking are local via Supermemory Local. Nothing leaves your machine on the default path.

Useful on first run

remindy seed infers standards from your repo, so recall returns real guidance before you correct anything.

On the “~15 tokens” number

That figure is the cost of a single compressed rule, the caveman line remindy injects. A real turn injects the handful of rules that match your task, and adds the two MCP tool definitions every MCP server carries. Recall itself makes no model call, it ranks on your machine, so the API cost on the default local path is zero.

Where to start

• New here? Run through the Quickstart to see the full loop in five minutes.
• Setting up a machine? See Install & setup.
• Want the mental model? Read How it works.