5 Ways AI Memory Transforms Your Development Workflow

> Explore this codebase and remember the key architectural decisions

[Claude explores and stores:]
- "This project uses Next.js 14 with App Router"
- "Database is PostgreSQL via Prisma ORM"
- "Auth is NextAuth with GitHub OAuth"
- "CSS is Tailwind with custom design tokens"

# Next day, new session:
> Add a new API endpoint for user preferences

[Claude already knows the tech stack, no re-explanation needed]

> Remember my coding preferences:
> - Always use functional components with hooks
> - TypeScript strict mode, no 'any' types
> - Error handling: try-catch with custom error classes
> - Tests: Vitest with React Testing Library
> - Comments: JSDoc for public APIs only

[Stored as Decision type memories]

# Future sessions automatically apply these preferences
> Create a new user profile component

[Claude generates functional component, strict TypeScript,
with Vitest tests—no reminders needed]

> The tests are failing with "Cannot find module '@/components/Button'"

[After investigation:]
> Remember: Path aliases require tsconfig.json paths to be mirrored
> in vitest.config.ts under resolve.alias

[Stored as Error type memory]

# Weeks later, similar issue:
> Why can't Vitest find my aliased imports?

[Claude recalls the previous error and solution immediately]

# Day 1:
> Remember: React Query's staleTime should be set higher for
> data that doesn't change often. We use 5 minutes for user profiles.

# Day 5:
> Remember: The Stripe webhook needs the raw body, so we disable
> body parsing for that route only.

# Day 12:
> Remember: PostgreSQL connection pool exhaustion happens when
> we don't close connections in serverless functions. Use
> connection poolers like PgBouncer.

# Day 30:
> Give me a summary of learnings about this project

[Claude retrieves all Learning type memories, providing
a knowledge base that grew organically over time]

# Session 1 (Monday):
> We're refactoring the auth system from session-based to JWT.
> Today we updated the login endpoint.

[Claude stores task context]

# Session 2 (Tuesday):
> What was I working on with the auth refactor?

[Claude recalls:]
"You were refactoring auth from session to JWT.
Login endpoint is updated. Remaining: logout, refresh tokens,
middleware updates."

# Session continues without re-explaining the whole context

# 1. Install the memory server
pip install shodh-memory

# 2. Add to Claude Code
claude mcp add @shodh/memory-mcp

# 3. Start using it
claude
> Remember that this project uses PostgreSQL with Prisma