
Addicted
4.2K posts










We are live on @ProductHunt The idea was simple when I first started. AI agents forget everything between sessions. That's broken. I wanted to fix it. Built the first version in a weekend. Scored 55% on LongMemEval. Not great. Kept shipping. Every week we'd push updates, run benchmarks, see what broke. Two weeks ago we rebuilt the entire retrieval pipeline and hit 81.3%, competitive with teams that raised millions. The whole thing has been built in public. There's a Telegram group where developers test Clude on their agents and tell us what's broken. That feedback loop is everything. Someone pointed out our entity cleanup was missing cascading deletes. Someone else caught that our recall API was silently capping results at 10. Real users finding real bugs. What's ready today: • Store and recall memories across sessions • 81.3% on LongMemEval, 1.96% hallucination rate • npm, Python, MCP - works with any agent framework • On-chain memory verification on Solana • 1.3M memories stored, 600+ agents We keep building and would love and appreciate the support producthunt.com/products/clude…




5 weeks. No whitepaper survives saltwater. W1 a mechanical contradiction in silicone. W2 the trap that catches what the ocean hides. W3 the nerve that tells you when you're sinking. W4 full throttle. taking the helm. open water. W5 not everything comes back from 100m. Deep end only 🌊




I love how @karpathy is always pulling from multiple different parts of computer science and technology. Always find the best tools of the job. Blockchains were made to synchronize untrusted parties in adversarial settings. Good tool for public AI work.















