Launching today
Torqon
Persistence Context for LLMs with inbuilt Token Reduction
6 followers
Persistence Context for LLMs with inbuilt Token Reduction
6 followers
Your AI forgets everything when a session ends. Torqon fixes that. It's a persistent memory layer for long-conversation LLM systems. Every session starts with context already loaded: projects, decisions, preferences, tech stack. After each message, it automatically extracts and stores new facts about you and your work. Works with Claude Code via hooks, and any MCP-compatible client. No more re-explaining yourself every session. 71.85% token reduction in heavy usage. Set up in under 2 minutes.
Free Options
Launch Team / Built With
I built this because I kept hitting the same wall.
Open a new Claude Code session. Spend the first twenty minutes typing out your stack, your conventions, what the project does, what you've already tried. Finally start coding. Close the session, open it the next day, do it all again.
And it gets worse. Claude compacts conversations as they grow. When that happens, most of what it "remembered" about you and your project is gone. It forgets your conventions, your decisions, things you told it an hour ago. You're back to square one mid-session, not just between sessions.
I tried workarounds. Massive system prompts, README files stuffed with context, markdown docs I'd paste at the start of every session. None of it stuck. I was spending more time re-explaining my own project than building it.
Torqon is the fix I made for myself. It's an MCP memory server that stores context across sessions and retrieves what's relevant when you need it. One env var to connect it. After that, your AI tools remember things between sessions without you doing anything.
The other thing, and I think this matters more than people realize: tokens. Torqon cuts token usage by up to 71% on heavy workloads because it pulls only what's relevant instead of stuffing everything into context at once. I looked at what else is out there. Nobody is doing this. The memory tools I found either ignore the token problem completely or make it worse. We built around it from the start.
Works with Claude Code, Cursor, anything MCP-compatible. There's also a community integration with Heym.
If you've found a better workaround for this problem, I'd actually like to hear it.
the automatic fact extraction after every message is a really thoughtful touch, it means nothing important slips through just because you forgot to manually save it