Four AI subscriptions, four separate conversation histories, four profiles of how you think. That's the reality for anyone bouncing between ChatGPT, Claude, Cursor, and Gemini daily.
The "Digital Twin Playbook," a new open-source project on GitHub, takes a straightforward approach to fixing this: put the memory layer in infrastructure you control, then connect every AI tool to it.
The setup runs a knowledge graph in Postgres (using Supabase's free tier) with pgvector, an extension that lets the database search by meaning rather than exact keywords. A small MCP server - MCP being the open protocol that lets AI assistants connect to external tools and data sources - reads from and writes to this graph. That server sits behind a gateway on a $6/month VPS, alongside Brave Search integration.
The practical result: when you explain your project structure to Claude, that context is available when you switch to ChatGPT or open Cursor. Your preferences, project details, and working patterns follow you instead of being locked inside each vendor's silo.
This scratches a real itch. Anyone who has re-explained their tech stack to a fresh Claude conversation, or watched ChatGPT forget a coding convention they have corrected five times, knows the friction. The current model where each AI provider builds its own memory system (ChatGPT's memory feature, Claude's project knowledge) creates vendor lock-in through accumulated context rather than through product quality.
The tradeoff is setup complexity. You need a Supabase account, a VPS, and enough comfort with Postgres and MCP configuration to wire everything together. This is a developer tool, not a consumer product. But the $6/month running cost is hard to argue with, and self-hosting means your conversation data stays out of yet another company's training pipeline.
The project is available on GitHub under the name digital-twin-playbook.