Own your AI's context and memories across every model and device
Hey HN, I pay for ChatGPT, Claude, Cursor, and use Gemini through work. Four vendors, four separate conversation histories, four profiles of how I think. None of them talk to each other. Switch providers and you start over. So I built a system where the memory is mine. I run a knowledge graph in Postgres (Supabase, free tier) with pgvector for semantic search. A small MCP server reads and writes to it. That server sits behind an MCP Gateway on a $6/month VPS, along with Brave Search and a GitHub server. TypingMind connects to the gateway as a BYOK client -- any model, any device, same…
What it does
In the maker’s words, at launch
Hey HN, I pay for ChatGPT, Claude, Cursor, and use Gemini through work. Four vendors, four separate conversation histories, four profiles of how I think. None of them talk to each other. Switch providers and you start over. So I built a system where the memory is mine. I run a knowledge graph in Postgres (Supabase, free tier) with pgvector for semantic search. A small MCP server reads and writes to it. That server sits behind an MCP Gateway on a $6/month VPS, along with Brave Search and a GitHub server. TypingMind connects to the gateway as a BYOK client -- any model, any device, same brain. https://github.com/kaspnilsson/digital-twin-playbook When I switch from Claude to GPT-5 to Gemini, the new model picks up where the old one left off. After three months of daily use, the AI knows my project architectures, my preferences, my side projects. I never re-explain any of it. The MCP server is MIT-licensed: https://github.com/kaspnilsson/mcp-memory-supabase The playbook walks through the full setup -- Supabase schema, VPS hardening, Caddy, systemd, the system prompt that makes tool routing work: https://github.com/kaspnilsson/digital-twin-playbook What it costs me: ~$45/month ($6 VPS + ~$36 API compute via OpenRouter + $3 amortized TypingMind license). More than a $20 subscription. But the $20 price is subsidized, and my data stays on my server. What does not work well: TypingMind is a PWA, not a native app. Voice is rough. iOS kills background processes on long tool chains. MCP config does not sync across devices. You are your own SRE. This is not a consumer product. If you want polish, use Claude.ai. If you want to own the context that makes your AI useful, this is how I did it!
Does the same job
all alternatives →
Second Brain for AIMay 2026 · github.com · ▲286Persistent memory for Claude, ChatGPT & Cursor. Free.
- COCore – open source memory graph for LLMs – shareable, user owned2025 · github.com · ▲112
I keep running in the same problem of each AI app “remembers” me in its own silo. ChatGPT knows my project details, Cursor forgets them, Claude starts from zero… so I end up re-explaining myself dozens of times a day across these apps. The deeper problem 1. Not portable – context is vendor-locked; nothing travels across tools. 2. Not relational – most memory systems store only the latest fact (“sticky notes”) with no history or provenance. 3. Not yours – your AI memory is sensitive first-party data, yet you have no control over where it lives or how it’s queried. Demo video:…
- IMI made the slowest, most expensive GPT2024 · ithy.com · ▲74
This is another one of my automate-my-life projects - I'm constantly asking the same question to different AIs since there's always the hope of getting a better answer somewhere else. Maybe ChatGPT's answer is too short, so I ask Perplexity. But I realize that's hallucinated, so I try Gemini. That answer sounds right, but I cross-reference with Claude just to make sure. This doesn't really apply to math/coding (where o1 or Gemini can probably one-shot an excellent response), but more to online search, where information is more fluid and there's no "right" search engine + text…
More ai this month
the category →
I trained a 125M-parameter transformer to autocomplete piano performances in real time (~108 notes/sec on an iPhone 15). The idea is basically GitHub Copilot or Tabnine, except instead of prompting it with code, you prompt it by playing a few notes on a MIDI piano. The model then continues what you played, entirely on-device. The app is free if anyone wants to try it. Happy to answer questions about the model, training, Core ML, or the many things that didn't work.
AI · 17d ago · simedw.com
Astute▲585Automate your B2B brand going viral, with new media creators
AI · 18d ago · company-app.joinastute.com
Hey HN, Henry from Cactus here! We previously released Cactus Needle, a 14MB agentic LLM for tool call, device use, and structured extraction for phones, wearables, smart homes, small robots and microcontrollers. We got really great feedback here, and have now incorporated the suggestions to release Needle 2. The whole model is a single 14MB binary that runs a full session in 28MB of RAM; 45m parameters at 2bit compression. Needle hits 500 tokens/sec decode speed on a Raspberry Pi 5, sits between 400-1,500 tokens/sec on VR devices like Meta Quest 3S and Apple Vision Pro, and ranges…
AI · 27d ago · cactuscompute.com
Launched alongside, March 2026
the whole month →
Switch from ChatGPT to Claude with import memory feature
AI · Mar 2026 · claude.com
