Design a Postgres Brain for Claude

Give Claude permanent memory by designing a structured knowledge base backed by PostgreSQL. Instead of re-explaining your business every session, Claude reads your "brain" and starts with full context.

The Prompt

Use this prompt in Claude Code or Cowork to have Claude design the schema for you:

// My tools:
[list your MCP connections]

// My business:
[describe what you do]

// My biggest gap:
[what context does your AI lose?]

--------------------------------------

Design my Postgres brain.

1. Schema (tables + columns)
2. Loading order
3. One sync per data source

// Start with clients.
// Ask me questions first.

How It Works

1. Schema Design

Claude designs a PostgreSQL schema tailored to your business. Typical tables:

• clients — name, industry, contact, notes, last interaction
• projects — status, deadlines, deliverables, client_id
• knowledge — topic, content, source, last_updated
• preferences — communication style, industry jargon, formatting rules
• history — past conversations, decisions, outcomes

2. Loading Order

Tables are loaded in dependency order so foreign keys resolve:

1. Core entities (clients, team members)
2. Relationships (projects, engagements)
3. Knowledge (industry data, preferences)
4. History (past interactions, decisions)

3. One Sync Per Data Source

Each external data source gets exactly one sync mechanism:

• CRM → clients table (daily sync)
• Project tool → projects table (hourly sync)
• Documents → knowledge table (on upload)
• Email → history table (weekly digest)

Why This Matters

• Claude starts every session with full business context
• No more "let me re-explain my business" prompts
• Decisions build on previous decisions
• Your AI gets smarter over time, not just faster

Getting Started

1. Run the prompt above in Claude Code
2. Let Claude ask you questions about your business
3. Review the proposed schema
4. Deploy to your Postgres instance
5. Set up sync scripts for your data sources

Best Practices

• Start with the clients table — it's the foundation
• Keep tables narrow (10–15 columns max)
• Use text columns for flexibility, add constraints later
• Include last_updated on every table for sync tracking
• Add a source column to track where data came from

Based on @angus.sewell's "Build my brain" Postgres prompt