Quickstart
Prerequisites
- Node.js 22 or higher
- A git repository (
git initif you don't have one)
Install
npm install -g mindragent
Initialize Mindr
Run mindr init in your project root:
cd my-project
mindr init
You will be prompted to choose a storage backend:
| Option | Description |
|---|---|
| Local SQLite | Zero-config. Stores memories in .mindr/mindr.sqlite. Best for solo or small team use. |
| Remembr cloud | Cross-machine persistence. Requires a Remembr base URL and API key. |
For most projects, SQLite is the right choice. You can migrate later with mindr migrate sqlite-to-remembr.
After init, Mindr:
- Creates
.mindr/config.toml - Installs a
post-commitgit hook that processes every commit automatically - Runs an initial convention scan across your codebase
✓ Mindr initialized
Backend: sqlite
Config: /my-project/.mindr/config.toml
Hook: /my-project/.git/hooks/post-commit
Verify
mindr status
Output shows backend type, hook status, and memory counts per type.
Connect your agent
Claude Code
Add to .claude/settings.json (project-level) or ~/.claude/settings.json (global):
{
"mcpServers": {
"mindr": {
"command": "mindr",
"args": ["serve"]
}
}
}
Restart Claude Code. On the next session, Claude will call mindr:get_context automatically and receive your stack, conventions, and recent decisions as structured context.
Other agents
See Cursor, Codex, Continue, Aider, OpenCode, and Windsurf.
Your first memories
Make a commit with a meaningful message. The post-commit hook processes it automatically:
git add .
git commit -m "Switch internal APIs from REST to tRPC for better type safety"
Mindr detects keywords like switch, migrate, refactor, dependency changes, and new directory structures, then stores a decision memory with a confidence score.
You can also store memories manually:
mindr remember "We use tRPC for all internal APIs" --type decision --module api
mindr remember "JWT expiry is 15m; refresh tokens live in Redis" --type decision --module auth
Generate AGENTS.md
After a few commits or manual memories, generate a structured snapshot:
mindr generate agents-md
This writes AGENTS.md to your project root, leading with an auto-detected
Commands section — build, test, run, and lint pulled from your package.json
scripts, Makefile, or language conventions (pytest, go test, cargo) — followed
by your stack, conventions, recent decisions, and active warnings. In a polyglot
monorepo, per-package commands are listed scoped by directory (e.g. cd server && pytest).
Prefer Claude Code's format, or want both?
mindr generate claude-md # CLAUDE.md
mindr generate --all # AGENTS.md + CLAUDE.md
Agents read these at session start. Re-run after major decisions to keep them fresh.
What agents receive
When an agent calls mindr:get_context, it receives a compact, priority-ordered brief
that is trimmed to fit a token budget:
=== MINDR CONTEXT ===
[WARNINGS]
FIXME `src/billing/invoice.ts:47` — retry logic
[RECENT DECISIONS]
2026-05-01 — switch internal APIs to tRPC [keyword]
2026-04-15 — JWT + Redis refresh tokens [keyword]
[COMMANDS]
Install: pnpm install
Build: pnpm build
Test: pnpm test
Lint: pnpm lint
[CONVENTIONS]
typescript: functionNames=camelCase(97%), classNames=PascalCase(100%), fileNames=kebab-case(89%)
[STACK OVERVIEW]
Languages: TypeScript
Stack: Express, PostgreSQL, tRPC
Hot modules: api (12), auth (7)
=== END CONTEXT ===
Next steps
- CLI Reference — all commands and flags
- SDK Reference — programmatic access from TypeScript
- Architecture — how Mindr works